xref: /linux/drivers/infiniband/hw/hns/hns_roce_hw_v2.c (revision 9e56ff53b4115875667760445b028357848b4748)
1 /*
2  * Copyright (c) 2016-2017 Hisilicon Limited.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #include <linux/acpi.h>
34 #include <linux/etherdevice.h>
35 #include <linux/interrupt.h>
36 #include <linux/iopoll.h>
37 #include <linux/kernel.h>
38 #include <linux/types.h>
39 #include <net/addrconf.h>
40 #include <rdma/ib_addr.h>
41 #include <rdma/ib_cache.h>
42 #include <rdma/ib_umem.h>
43 #include <rdma/uverbs_ioctl.h>
44 
45 #include "hnae3.h"
46 #include "hns_roce_common.h"
47 #include "hns_roce_device.h"
48 #include "hns_roce_cmd.h"
49 #include "hns_roce_hem.h"
50 #include "hns_roce_hw_v2.h"
51 
52 enum {
53 	CMD_RST_PRC_OTHERS,
54 	CMD_RST_PRC_SUCCESS,
55 	CMD_RST_PRC_EBUSY,
56 };
57 
58 enum ecc_resource_type {
59 	ECC_RESOURCE_QPC,
60 	ECC_RESOURCE_CQC,
61 	ECC_RESOURCE_MPT,
62 	ECC_RESOURCE_SRQC,
63 	ECC_RESOURCE_GMV,
64 	ECC_RESOURCE_QPC_TIMER,
65 	ECC_RESOURCE_CQC_TIMER,
66 	ECC_RESOURCE_SCCC,
67 	ECC_RESOURCE_COUNT,
68 };
69 
70 static const struct {
71 	const char *name;
72 	u8 read_bt0_op;
73 	u8 write_bt0_op;
74 } fmea_ram_res[] = {
75 	{ "ECC_RESOURCE_QPC",
76 	  HNS_ROCE_CMD_READ_QPC_BT0, HNS_ROCE_CMD_WRITE_QPC_BT0 },
77 	{ "ECC_RESOURCE_CQC",
78 	  HNS_ROCE_CMD_READ_CQC_BT0, HNS_ROCE_CMD_WRITE_CQC_BT0 },
79 	{ "ECC_RESOURCE_MPT",
80 	  HNS_ROCE_CMD_READ_MPT_BT0, HNS_ROCE_CMD_WRITE_MPT_BT0 },
81 	{ "ECC_RESOURCE_SRQC",
82 	  HNS_ROCE_CMD_READ_SRQC_BT0, HNS_ROCE_CMD_WRITE_SRQC_BT0 },
83 	/* ECC_RESOURCE_GMV is handled by cmdq, not mailbox */
84 	{ "ECC_RESOURCE_GMV",
85 	  0, 0 },
86 	{ "ECC_RESOURCE_QPC_TIMER",
87 	  HNS_ROCE_CMD_READ_QPC_TIMER_BT0, HNS_ROCE_CMD_WRITE_QPC_TIMER_BT0 },
88 	{ "ECC_RESOURCE_CQC_TIMER",
89 	  HNS_ROCE_CMD_READ_CQC_TIMER_BT0, HNS_ROCE_CMD_WRITE_CQC_TIMER_BT0 },
90 	{ "ECC_RESOURCE_SCCC",
91 	  HNS_ROCE_CMD_READ_SCCC_BT0, HNS_ROCE_CMD_WRITE_SCCC_BT0 },
92 };
93 
94 static inline void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
95 				   struct ib_sge *sg)
96 {
97 	dseg->lkey = cpu_to_le32(sg->lkey);
98 	dseg->addr = cpu_to_le64(sg->addr);
99 	dseg->len  = cpu_to_le32(sg->length);
100 }
101 
102 /*
103  * mapped-value = 1 + real-value
104  * The hns wr opcode real value is start from 0, In order to distinguish between
105  * initialized and uninitialized map values, we plus 1 to the actual value when
106  * defining the mapping, so that the validity can be identified by checking the
107  * mapped value is greater than 0.
108  */
109 #define HR_OPC_MAP(ib_key, hr_key) \
110 		[IB_WR_ ## ib_key] = 1 + HNS_ROCE_V2_WQE_OP_ ## hr_key
111 
112 static const u32 hns_roce_op_code[] = {
113 	HR_OPC_MAP(RDMA_WRITE,			RDMA_WRITE),
114 	HR_OPC_MAP(RDMA_WRITE_WITH_IMM,		RDMA_WRITE_WITH_IMM),
115 	HR_OPC_MAP(SEND,			SEND),
116 	HR_OPC_MAP(SEND_WITH_IMM,		SEND_WITH_IMM),
117 	HR_OPC_MAP(RDMA_READ,			RDMA_READ),
118 	HR_OPC_MAP(ATOMIC_CMP_AND_SWP,		ATOM_CMP_AND_SWAP),
119 	HR_OPC_MAP(ATOMIC_FETCH_AND_ADD,	ATOM_FETCH_AND_ADD),
120 	HR_OPC_MAP(SEND_WITH_INV,		SEND_WITH_INV),
121 	HR_OPC_MAP(MASKED_ATOMIC_CMP_AND_SWP,	ATOM_MSK_CMP_AND_SWAP),
122 	HR_OPC_MAP(MASKED_ATOMIC_FETCH_AND_ADD,	ATOM_MSK_FETCH_AND_ADD),
123 	HR_OPC_MAP(REG_MR,			FAST_REG_PMR),
124 };
125 
126 static u32 to_hr_opcode(u32 ib_opcode)
127 {
128 	if (ib_opcode >= ARRAY_SIZE(hns_roce_op_code))
129 		return HNS_ROCE_V2_WQE_OP_MASK;
130 
131 	return hns_roce_op_code[ib_opcode] ? hns_roce_op_code[ib_opcode] - 1 :
132 					     HNS_ROCE_V2_WQE_OP_MASK;
133 }
134 
135 static void set_frmr_seg(struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
136 			 const struct ib_reg_wr *wr)
137 {
138 	struct hns_roce_wqe_frmr_seg *fseg =
139 		(void *)rc_sq_wqe + sizeof(struct hns_roce_v2_rc_send_wqe);
140 	struct hns_roce_mr *mr = to_hr_mr(wr->mr);
141 	u64 pbl_ba;
142 
143 	/* use ib_access_flags */
144 	hr_reg_write_bool(fseg, FRMR_BIND_EN, wr->access & IB_ACCESS_MW_BIND);
145 	hr_reg_write_bool(fseg, FRMR_ATOMIC,
146 			  wr->access & IB_ACCESS_REMOTE_ATOMIC);
147 	hr_reg_write_bool(fseg, FRMR_RR, wr->access & IB_ACCESS_REMOTE_READ);
148 	hr_reg_write_bool(fseg, FRMR_RW, wr->access & IB_ACCESS_REMOTE_WRITE);
149 	hr_reg_write_bool(fseg, FRMR_LW, wr->access & IB_ACCESS_LOCAL_WRITE);
150 
151 	/* Data structure reuse may lead to confusion */
152 	pbl_ba = mr->pbl_mtr.hem_cfg.root_ba;
153 	rc_sq_wqe->msg_len = cpu_to_le32(lower_32_bits(pbl_ba));
154 	rc_sq_wqe->inv_key = cpu_to_le32(upper_32_bits(pbl_ba));
155 
156 	rc_sq_wqe->byte_16 = cpu_to_le32(wr->mr->length & 0xffffffff);
157 	rc_sq_wqe->byte_20 = cpu_to_le32(wr->mr->length >> 32);
158 	rc_sq_wqe->rkey = cpu_to_le32(wr->key);
159 	rc_sq_wqe->va = cpu_to_le64(wr->mr->iova);
160 
161 	hr_reg_write(fseg, FRMR_PBL_SIZE, mr->npages);
162 	hr_reg_write(fseg, FRMR_PBL_BUF_PG_SZ,
163 		     to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.buf_pg_shift));
164 	hr_reg_clear(fseg, FRMR_BLK_MODE);
165 }
166 
167 static void set_atomic_seg(const struct ib_send_wr *wr,
168 			   struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
169 			   unsigned int valid_num_sge)
170 {
171 	struct hns_roce_v2_wqe_data_seg *dseg =
172 		(void *)rc_sq_wqe + sizeof(struct hns_roce_v2_rc_send_wqe);
173 	struct hns_roce_wqe_atomic_seg *aseg =
174 		(void *)dseg + sizeof(struct hns_roce_v2_wqe_data_seg);
175 
176 	set_data_seg_v2(dseg, wr->sg_list);
177 
178 	if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
179 		aseg->fetchadd_swap_data = cpu_to_le64(atomic_wr(wr)->swap);
180 		aseg->cmp_data = cpu_to_le64(atomic_wr(wr)->compare_add);
181 	} else {
182 		aseg->fetchadd_swap_data =
183 			cpu_to_le64(atomic_wr(wr)->compare_add);
184 		aseg->cmp_data = 0;
185 	}
186 
187 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_SGE_NUM, valid_num_sge);
188 }
189 
190 static int fill_ext_sge_inl_data(struct hns_roce_qp *qp,
191 				 const struct ib_send_wr *wr,
192 				 unsigned int *sge_idx, u32 msg_len)
193 {
194 	struct ib_device *ibdev = &(to_hr_dev(qp->ibqp.device))->ib_dev;
195 	unsigned int left_len_in_pg;
196 	unsigned int idx = *sge_idx;
197 	unsigned int i = 0;
198 	unsigned int len;
199 	void *addr;
200 	void *dseg;
201 
202 	if (msg_len > qp->sq.ext_sge_cnt * HNS_ROCE_SGE_SIZE) {
203 		ibdev_err(ibdev,
204 			  "no enough extended sge space for inline data.\n");
205 		return -EINVAL;
206 	}
207 
208 	dseg = hns_roce_get_extend_sge(qp, idx & (qp->sge.sge_cnt - 1));
209 	left_len_in_pg = hr_hw_page_align((uintptr_t)dseg) - (uintptr_t)dseg;
210 	len = wr->sg_list[0].length;
211 	addr = (void *)(unsigned long)(wr->sg_list[0].addr);
212 
213 	/* When copying data to extended sge space, the left length in page may
214 	 * not long enough for current user's sge. So the data should be
215 	 * splited into several parts, one in the first page, and the others in
216 	 * the subsequent pages.
217 	 */
218 	while (1) {
219 		if (len <= left_len_in_pg) {
220 			memcpy(dseg, addr, len);
221 
222 			idx += len / HNS_ROCE_SGE_SIZE;
223 
224 			i++;
225 			if (i >= wr->num_sge)
226 				break;
227 
228 			left_len_in_pg -= len;
229 			len = wr->sg_list[i].length;
230 			addr = (void *)(unsigned long)(wr->sg_list[i].addr);
231 			dseg += len;
232 		} else {
233 			memcpy(dseg, addr, left_len_in_pg);
234 
235 			len -= left_len_in_pg;
236 			addr += left_len_in_pg;
237 			idx += left_len_in_pg / HNS_ROCE_SGE_SIZE;
238 			dseg = hns_roce_get_extend_sge(qp,
239 						idx & (qp->sge.sge_cnt - 1));
240 			left_len_in_pg = 1 << HNS_HW_PAGE_SHIFT;
241 		}
242 	}
243 
244 	*sge_idx = idx;
245 
246 	return 0;
247 }
248 
249 static void set_extend_sge(struct hns_roce_qp *qp, struct ib_sge *sge,
250 			   unsigned int *sge_ind, unsigned int cnt)
251 {
252 	struct hns_roce_v2_wqe_data_seg *dseg;
253 	unsigned int idx = *sge_ind;
254 
255 	while (cnt > 0) {
256 		dseg = hns_roce_get_extend_sge(qp, idx & (qp->sge.sge_cnt - 1));
257 		if (likely(sge->length)) {
258 			set_data_seg_v2(dseg, sge);
259 			idx++;
260 			cnt--;
261 		}
262 		sge++;
263 	}
264 
265 	*sge_ind = idx;
266 }
267 
268 static bool check_inl_data_len(struct hns_roce_qp *qp, unsigned int len)
269 {
270 	struct hns_roce_dev *hr_dev = to_hr_dev(qp->ibqp.device);
271 	int mtu = ib_mtu_enum_to_int(qp->path_mtu);
272 
273 	if (mtu < 0 || len > qp->max_inline_data || len > mtu) {
274 		ibdev_err(&hr_dev->ib_dev,
275 			  "invalid length of data, data len = %u, max inline len = %u, path mtu = %d.\n",
276 			  len, qp->max_inline_data, mtu);
277 		return false;
278 	}
279 
280 	return true;
281 }
282 
283 static int set_rc_inl(struct hns_roce_qp *qp, const struct ib_send_wr *wr,
284 		      struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
285 		      unsigned int *sge_idx)
286 {
287 	struct hns_roce_dev *hr_dev = to_hr_dev(qp->ibqp.device);
288 	u32 msg_len = le32_to_cpu(rc_sq_wqe->msg_len);
289 	struct ib_device *ibdev = &hr_dev->ib_dev;
290 	unsigned int curr_idx = *sge_idx;
291 	void *dseg = rc_sq_wqe;
292 	unsigned int i;
293 	int ret;
294 
295 	if (unlikely(wr->opcode == IB_WR_RDMA_READ)) {
296 		ibdev_err(ibdev, "invalid inline parameters!\n");
297 		return -EINVAL;
298 	}
299 
300 	if (!check_inl_data_len(qp, msg_len))
301 		return -EINVAL;
302 
303 	dseg += sizeof(struct hns_roce_v2_rc_send_wqe);
304 
305 	if (msg_len <= HNS_ROCE_V2_MAX_RC_INL_INN_SZ) {
306 		hr_reg_clear(rc_sq_wqe, RC_SEND_WQE_INL_TYPE);
307 
308 		for (i = 0; i < wr->num_sge; i++) {
309 			memcpy(dseg, ((void *)wr->sg_list[i].addr),
310 			       wr->sg_list[i].length);
311 			dseg += wr->sg_list[i].length;
312 		}
313 	} else {
314 		hr_reg_enable(rc_sq_wqe, RC_SEND_WQE_INL_TYPE);
315 
316 		ret = fill_ext_sge_inl_data(qp, wr, &curr_idx, msg_len);
317 		if (ret)
318 			return ret;
319 
320 		hr_reg_write(rc_sq_wqe, RC_SEND_WQE_SGE_NUM, curr_idx - *sge_idx);
321 	}
322 
323 	*sge_idx = curr_idx;
324 
325 	return 0;
326 }
327 
328 static int set_rwqe_data_seg(struct ib_qp *ibqp, const struct ib_send_wr *wr,
329 			     struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
330 			     unsigned int *sge_ind,
331 			     unsigned int valid_num_sge)
332 {
333 	struct hns_roce_v2_wqe_data_seg *dseg =
334 		(void *)rc_sq_wqe + sizeof(struct hns_roce_v2_rc_send_wqe);
335 	struct hns_roce_qp *qp = to_hr_qp(ibqp);
336 	int j = 0;
337 	int i;
338 
339 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_MSG_START_SGE_IDX,
340 		     (*sge_ind) & (qp->sge.sge_cnt - 1));
341 
342 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_INLINE,
343 		     !!(wr->send_flags & IB_SEND_INLINE));
344 	if (wr->send_flags & IB_SEND_INLINE)
345 		return set_rc_inl(qp, wr, rc_sq_wqe, sge_ind);
346 
347 	if (valid_num_sge <= HNS_ROCE_SGE_IN_WQE) {
348 		for (i = 0; i < wr->num_sge; i++) {
349 			if (likely(wr->sg_list[i].length)) {
350 				set_data_seg_v2(dseg, wr->sg_list + i);
351 				dseg++;
352 			}
353 		}
354 	} else {
355 		for (i = 0; i < wr->num_sge && j < HNS_ROCE_SGE_IN_WQE; i++) {
356 			if (likely(wr->sg_list[i].length)) {
357 				set_data_seg_v2(dseg, wr->sg_list + i);
358 				dseg++;
359 				j++;
360 			}
361 		}
362 
363 		set_extend_sge(qp, wr->sg_list + i, sge_ind,
364 			       valid_num_sge - HNS_ROCE_SGE_IN_WQE);
365 	}
366 
367 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_SGE_NUM, valid_num_sge);
368 
369 	return 0;
370 }
371 
372 static int check_send_valid(struct hns_roce_dev *hr_dev,
373 			    struct hns_roce_qp *hr_qp)
374 {
375 	struct ib_device *ibdev = &hr_dev->ib_dev;
376 
377 	if (unlikely(hr_qp->state == IB_QPS_RESET ||
378 		     hr_qp->state == IB_QPS_INIT ||
379 		     hr_qp->state == IB_QPS_RTR)) {
380 		ibdev_err(ibdev, "failed to post WQE, QP state %u!\n",
381 			  hr_qp->state);
382 		return -EINVAL;
383 	} else if (unlikely(hr_dev->state >= HNS_ROCE_DEVICE_STATE_RST_DOWN)) {
384 		ibdev_err(ibdev, "failed to post WQE, dev state %d!\n",
385 			  hr_dev->state);
386 		return -EIO;
387 	}
388 
389 	return 0;
390 }
391 
392 static unsigned int calc_wr_sge_num(const struct ib_send_wr *wr,
393 				    unsigned int *sge_len)
394 {
395 	unsigned int valid_num = 0;
396 	unsigned int len = 0;
397 	int i;
398 
399 	for (i = 0; i < wr->num_sge; i++) {
400 		if (likely(wr->sg_list[i].length)) {
401 			len += wr->sg_list[i].length;
402 			valid_num++;
403 		}
404 	}
405 
406 	*sge_len = len;
407 	return valid_num;
408 }
409 
410 static __le32 get_immtdata(const struct ib_send_wr *wr)
411 {
412 	switch (wr->opcode) {
413 	case IB_WR_SEND_WITH_IMM:
414 	case IB_WR_RDMA_WRITE_WITH_IMM:
415 		return cpu_to_le32(be32_to_cpu(wr->ex.imm_data));
416 	default:
417 		return 0;
418 	}
419 }
420 
421 static int set_ud_opcode(struct hns_roce_v2_ud_send_wqe *ud_sq_wqe,
422 			 const struct ib_send_wr *wr)
423 {
424 	u32 ib_op = wr->opcode;
425 
426 	if (ib_op != IB_WR_SEND && ib_op != IB_WR_SEND_WITH_IMM)
427 		return -EINVAL;
428 
429 	ud_sq_wqe->immtdata = get_immtdata(wr);
430 
431 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_OPCODE, to_hr_opcode(ib_op));
432 
433 	return 0;
434 }
435 
436 static int fill_ud_av(struct hns_roce_v2_ud_send_wqe *ud_sq_wqe,
437 		      struct hns_roce_ah *ah)
438 {
439 	struct ib_device *ib_dev = ah->ibah.device;
440 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
441 
442 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_UDPSPN, ah->av.udp_sport);
443 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_HOPLIMIT, ah->av.hop_limit);
444 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_TCLASS, ah->av.tclass);
445 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_FLOW_LABEL, ah->av.flowlabel);
446 
447 	if (WARN_ON(ah->av.sl > MAX_SERVICE_LEVEL))
448 		return -EINVAL;
449 
450 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_SL, ah->av.sl);
451 
452 	ud_sq_wqe->sgid_index = ah->av.gid_index;
453 
454 	memcpy(ud_sq_wqe->dmac, ah->av.mac, ETH_ALEN);
455 	memcpy(ud_sq_wqe->dgid, ah->av.dgid, GID_LEN_V2);
456 
457 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
458 		return 0;
459 
460 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_VLAN_EN, ah->av.vlan_en);
461 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_VLAN, ah->av.vlan_id);
462 
463 	return 0;
464 }
465 
466 static inline int set_ud_wqe(struct hns_roce_qp *qp,
467 			     const struct ib_send_wr *wr,
468 			     void *wqe, unsigned int *sge_idx,
469 			     unsigned int owner_bit)
470 {
471 	struct hns_roce_ah *ah = to_hr_ah(ud_wr(wr)->ah);
472 	struct hns_roce_v2_ud_send_wqe *ud_sq_wqe = wqe;
473 	unsigned int curr_idx = *sge_idx;
474 	unsigned int valid_num_sge;
475 	u32 msg_len = 0;
476 	int ret;
477 
478 	valid_num_sge = calc_wr_sge_num(wr, &msg_len);
479 
480 	ret = set_ud_opcode(ud_sq_wqe, wr);
481 	if (WARN_ON(ret))
482 		return ret;
483 
484 	ud_sq_wqe->msg_len = cpu_to_le32(msg_len);
485 
486 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_CQE,
487 		     !!(wr->send_flags & IB_SEND_SIGNALED));
488 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_SE,
489 		     !!(wr->send_flags & IB_SEND_SOLICITED));
490 
491 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_PD, to_hr_pd(qp->ibqp.pd)->pdn);
492 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_SGE_NUM, valid_num_sge);
493 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_MSG_START_SGE_IDX,
494 		     curr_idx & (qp->sge.sge_cnt - 1));
495 
496 	ud_sq_wqe->qkey = cpu_to_le32(ud_wr(wr)->remote_qkey & 0x80000000 ?
497 			  qp->qkey : ud_wr(wr)->remote_qkey);
498 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_DQPN, ud_wr(wr)->remote_qpn);
499 
500 	ret = fill_ud_av(ud_sq_wqe, ah);
501 	if (ret)
502 		return ret;
503 
504 	qp->sl = to_hr_ah(ud_wr(wr)->ah)->av.sl;
505 
506 	set_extend_sge(qp, wr->sg_list, &curr_idx, valid_num_sge);
507 
508 	/*
509 	 * The pipeline can sequentially post all valid WQEs into WQ buffer,
510 	 * including new WQEs waiting for the doorbell to update the PI again.
511 	 * Therefore, the owner bit of WQE MUST be updated after all fields
512 	 * and extSGEs have been written into DDR instead of cache.
513 	 */
514 	if (qp->en_flags & HNS_ROCE_QP_CAP_OWNER_DB)
515 		dma_wmb();
516 
517 	*sge_idx = curr_idx;
518 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_OWNER, owner_bit);
519 
520 	return 0;
521 }
522 
523 static int set_rc_opcode(struct hns_roce_dev *hr_dev,
524 			 struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
525 			 const struct ib_send_wr *wr)
526 {
527 	u32 ib_op = wr->opcode;
528 	int ret = 0;
529 
530 	rc_sq_wqe->immtdata = get_immtdata(wr);
531 
532 	switch (ib_op) {
533 	case IB_WR_RDMA_READ:
534 	case IB_WR_RDMA_WRITE:
535 	case IB_WR_RDMA_WRITE_WITH_IMM:
536 		rc_sq_wqe->rkey = cpu_to_le32(rdma_wr(wr)->rkey);
537 		rc_sq_wqe->va = cpu_to_le64(rdma_wr(wr)->remote_addr);
538 		break;
539 	case IB_WR_SEND:
540 	case IB_WR_SEND_WITH_IMM:
541 		break;
542 	case IB_WR_ATOMIC_CMP_AND_SWP:
543 	case IB_WR_ATOMIC_FETCH_AND_ADD:
544 		rc_sq_wqe->rkey = cpu_to_le32(atomic_wr(wr)->rkey);
545 		rc_sq_wqe->va = cpu_to_le64(atomic_wr(wr)->remote_addr);
546 		break;
547 	case IB_WR_REG_MR:
548 		if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
549 			set_frmr_seg(rc_sq_wqe, reg_wr(wr));
550 		else
551 			ret = -EOPNOTSUPP;
552 		break;
553 	case IB_WR_SEND_WITH_INV:
554 		rc_sq_wqe->inv_key = cpu_to_le32(wr->ex.invalidate_rkey);
555 		break;
556 	default:
557 		ret = -EINVAL;
558 	}
559 
560 	if (unlikely(ret))
561 		return ret;
562 
563 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_OPCODE, to_hr_opcode(ib_op));
564 
565 	return ret;
566 }
567 
568 static inline int set_rc_wqe(struct hns_roce_qp *qp,
569 			     const struct ib_send_wr *wr,
570 			     void *wqe, unsigned int *sge_idx,
571 			     unsigned int owner_bit)
572 {
573 	struct hns_roce_dev *hr_dev = to_hr_dev(qp->ibqp.device);
574 	struct hns_roce_v2_rc_send_wqe *rc_sq_wqe = wqe;
575 	unsigned int curr_idx = *sge_idx;
576 	unsigned int valid_num_sge;
577 	u32 msg_len = 0;
578 	int ret;
579 
580 	valid_num_sge = calc_wr_sge_num(wr, &msg_len);
581 
582 	rc_sq_wqe->msg_len = cpu_to_le32(msg_len);
583 
584 	ret = set_rc_opcode(hr_dev, rc_sq_wqe, wr);
585 	if (WARN_ON(ret))
586 		return ret;
587 
588 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_FENCE,
589 		     (wr->send_flags & IB_SEND_FENCE) ? 1 : 0);
590 
591 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_SE,
592 		     (wr->send_flags & IB_SEND_SOLICITED) ? 1 : 0);
593 
594 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_CQE,
595 		     (wr->send_flags & IB_SEND_SIGNALED) ? 1 : 0);
596 
597 	if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
598 	    wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
599 		set_atomic_seg(wr, rc_sq_wqe, valid_num_sge);
600 	else if (wr->opcode != IB_WR_REG_MR)
601 		ret = set_rwqe_data_seg(&qp->ibqp, wr, rc_sq_wqe,
602 					&curr_idx, valid_num_sge);
603 
604 	/*
605 	 * The pipeline can sequentially post all valid WQEs into WQ buffer,
606 	 * including new WQEs waiting for the doorbell to update the PI again.
607 	 * Therefore, the owner bit of WQE MUST be updated after all fields
608 	 * and extSGEs have been written into DDR instead of cache.
609 	 */
610 	if (qp->en_flags & HNS_ROCE_QP_CAP_OWNER_DB)
611 		dma_wmb();
612 
613 	*sge_idx = curr_idx;
614 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_OWNER, owner_bit);
615 
616 	return ret;
617 }
618 
619 static inline void update_sq_db(struct hns_roce_dev *hr_dev,
620 				struct hns_roce_qp *qp)
621 {
622 	if (unlikely(qp->state == IB_QPS_ERR)) {
623 		flush_cqe(hr_dev, qp);
624 	} else {
625 		struct hns_roce_v2_db sq_db = {};
626 
627 		hr_reg_write(&sq_db, DB_TAG, qp->qpn);
628 		hr_reg_write(&sq_db, DB_CMD, HNS_ROCE_V2_SQ_DB);
629 		hr_reg_write(&sq_db, DB_PI, qp->sq.head);
630 		hr_reg_write(&sq_db, DB_SL, qp->sl);
631 
632 		hns_roce_write64(hr_dev, (__le32 *)&sq_db, qp->sq.db_reg);
633 	}
634 }
635 
636 static inline void update_rq_db(struct hns_roce_dev *hr_dev,
637 				struct hns_roce_qp *qp)
638 {
639 	if (unlikely(qp->state == IB_QPS_ERR)) {
640 		flush_cqe(hr_dev, qp);
641 	} else {
642 		if (likely(qp->en_flags & HNS_ROCE_QP_CAP_RQ_RECORD_DB)) {
643 			*qp->rdb.db_record =
644 					qp->rq.head & V2_DB_PRODUCER_IDX_M;
645 		} else {
646 			struct hns_roce_v2_db rq_db = {};
647 
648 			hr_reg_write(&rq_db, DB_TAG, qp->qpn);
649 			hr_reg_write(&rq_db, DB_CMD, HNS_ROCE_V2_RQ_DB);
650 			hr_reg_write(&rq_db, DB_PI, qp->rq.head);
651 
652 			hns_roce_write64(hr_dev, (__le32 *)&rq_db,
653 					 qp->rq.db_reg);
654 		}
655 	}
656 }
657 
658 static void hns_roce_write512(struct hns_roce_dev *hr_dev, u64 *val,
659 			      u64 __iomem *dest)
660 {
661 #define HNS_ROCE_WRITE_TIMES 8
662 	struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv;
663 	struct hnae3_handle *handle = priv->handle;
664 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
665 	int i;
666 
667 	if (!hr_dev->dis_db && !ops->get_hw_reset_stat(handle))
668 		for (i = 0; i < HNS_ROCE_WRITE_TIMES; i++)
669 			writeq_relaxed(*(val + i), dest + i);
670 }
671 
672 static void write_dwqe(struct hns_roce_dev *hr_dev, struct hns_roce_qp *qp,
673 		       void *wqe)
674 {
675 #define HNS_ROCE_SL_SHIFT 2
676 	struct hns_roce_v2_rc_send_wqe *rc_sq_wqe = wqe;
677 
678 	/* All kinds of DirectWQE have the same header field layout */
679 	hr_reg_enable(rc_sq_wqe, RC_SEND_WQE_FLAG);
680 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_DB_SL_L, qp->sl);
681 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_DB_SL_H,
682 		     qp->sl >> HNS_ROCE_SL_SHIFT);
683 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_WQE_INDEX, qp->sq.head);
684 
685 	hns_roce_write512(hr_dev, wqe, qp->sq.db_reg);
686 }
687 
688 static int hns_roce_v2_post_send(struct ib_qp *ibqp,
689 				 const struct ib_send_wr *wr,
690 				 const struct ib_send_wr **bad_wr)
691 {
692 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
693 	struct ib_device *ibdev = &hr_dev->ib_dev;
694 	struct hns_roce_qp *qp = to_hr_qp(ibqp);
695 	unsigned long flags = 0;
696 	unsigned int owner_bit;
697 	unsigned int sge_idx;
698 	unsigned int wqe_idx;
699 	void *wqe = NULL;
700 	u32 nreq;
701 	int ret;
702 
703 	spin_lock_irqsave(&qp->sq.lock, flags);
704 
705 	ret = check_send_valid(hr_dev, qp);
706 	if (unlikely(ret)) {
707 		*bad_wr = wr;
708 		nreq = 0;
709 		goto out;
710 	}
711 
712 	sge_idx = qp->next_sge;
713 
714 	for (nreq = 0; wr; ++nreq, wr = wr->next) {
715 		if (hns_roce_wq_overflow(&qp->sq, nreq, qp->ibqp.send_cq)) {
716 			ret = -ENOMEM;
717 			*bad_wr = wr;
718 			goto out;
719 		}
720 
721 		wqe_idx = (qp->sq.head + nreq) & (qp->sq.wqe_cnt - 1);
722 
723 		if (unlikely(wr->num_sge > qp->sq.max_gs)) {
724 			ibdev_err(ibdev, "num_sge = %d > qp->sq.max_gs = %u.\n",
725 				  wr->num_sge, qp->sq.max_gs);
726 			ret = -EINVAL;
727 			*bad_wr = wr;
728 			goto out;
729 		}
730 
731 		wqe = hns_roce_get_send_wqe(qp, wqe_idx);
732 		qp->sq.wrid[wqe_idx] = wr->wr_id;
733 		owner_bit =
734 		       ~(((qp->sq.head + nreq) >> ilog2(qp->sq.wqe_cnt)) & 0x1);
735 
736 		/* Corresponding to the QP type, wqe process separately */
737 		if (ibqp->qp_type == IB_QPT_RC)
738 			ret = set_rc_wqe(qp, wr, wqe, &sge_idx, owner_bit);
739 		else
740 			ret = set_ud_wqe(qp, wr, wqe, &sge_idx, owner_bit);
741 
742 		if (unlikely(ret)) {
743 			*bad_wr = wr;
744 			goto out;
745 		}
746 	}
747 
748 out:
749 	if (likely(nreq)) {
750 		qp->sq.head += nreq;
751 		qp->next_sge = sge_idx;
752 
753 		if (nreq == 1 && !ret &&
754 		    (qp->en_flags & HNS_ROCE_QP_CAP_DIRECT_WQE))
755 			write_dwqe(hr_dev, qp, wqe);
756 		else
757 			update_sq_db(hr_dev, qp);
758 	}
759 
760 	spin_unlock_irqrestore(&qp->sq.lock, flags);
761 
762 	return ret;
763 }
764 
765 static int check_recv_valid(struct hns_roce_dev *hr_dev,
766 			    struct hns_roce_qp *hr_qp)
767 {
768 	if (unlikely(hr_dev->state >= HNS_ROCE_DEVICE_STATE_RST_DOWN))
769 		return -EIO;
770 
771 	if (hr_qp->state == IB_QPS_RESET)
772 		return -EINVAL;
773 
774 	return 0;
775 }
776 
777 static void fill_recv_sge_to_wqe(const struct ib_recv_wr *wr, void *wqe,
778 				 u32 max_sge, bool rsv)
779 {
780 	struct hns_roce_v2_wqe_data_seg *dseg = wqe;
781 	u32 i, cnt;
782 
783 	for (i = 0, cnt = 0; i < wr->num_sge; i++) {
784 		/* Skip zero-length sge */
785 		if (!wr->sg_list[i].length)
786 			continue;
787 		set_data_seg_v2(dseg + cnt, wr->sg_list + i);
788 		cnt++;
789 	}
790 
791 	/* Fill a reserved sge to make hw stop reading remaining segments */
792 	if (rsv) {
793 		dseg[cnt].lkey = cpu_to_le32(HNS_ROCE_INVALID_LKEY);
794 		dseg[cnt].addr = 0;
795 		dseg[cnt].len = cpu_to_le32(HNS_ROCE_INVALID_SGE_LENGTH);
796 	} else {
797 		/* Clear remaining segments to make ROCEE ignore sges */
798 		if (cnt < max_sge)
799 			memset(dseg + cnt, 0,
800 			       (max_sge - cnt) * HNS_ROCE_SGE_SIZE);
801 	}
802 }
803 
804 static void fill_rq_wqe(struct hns_roce_qp *hr_qp, const struct ib_recv_wr *wr,
805 			u32 wqe_idx, u32 max_sge)
806 {
807 	void *wqe = NULL;
808 
809 	wqe = hns_roce_get_recv_wqe(hr_qp, wqe_idx);
810 	fill_recv_sge_to_wqe(wr, wqe, max_sge, hr_qp->rq.rsv_sge);
811 }
812 
813 static int hns_roce_v2_post_recv(struct ib_qp *ibqp,
814 				 const struct ib_recv_wr *wr,
815 				 const struct ib_recv_wr **bad_wr)
816 {
817 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
818 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
819 	struct ib_device *ibdev = &hr_dev->ib_dev;
820 	u32 wqe_idx, nreq, max_sge;
821 	unsigned long flags;
822 	int ret;
823 
824 	spin_lock_irqsave(&hr_qp->rq.lock, flags);
825 
826 	ret = check_recv_valid(hr_dev, hr_qp);
827 	if (unlikely(ret)) {
828 		*bad_wr = wr;
829 		nreq = 0;
830 		goto out;
831 	}
832 
833 	max_sge = hr_qp->rq.max_gs - hr_qp->rq.rsv_sge;
834 	for (nreq = 0; wr; ++nreq, wr = wr->next) {
835 		if (unlikely(hns_roce_wq_overflow(&hr_qp->rq, nreq,
836 						  hr_qp->ibqp.recv_cq))) {
837 			ret = -ENOMEM;
838 			*bad_wr = wr;
839 			goto out;
840 		}
841 
842 		if (unlikely(wr->num_sge > max_sge)) {
843 			ibdev_err(ibdev, "num_sge = %d >= max_sge = %u.\n",
844 				  wr->num_sge, max_sge);
845 			ret = -EINVAL;
846 			*bad_wr = wr;
847 			goto out;
848 		}
849 
850 		wqe_idx = (hr_qp->rq.head + nreq) & (hr_qp->rq.wqe_cnt - 1);
851 		fill_rq_wqe(hr_qp, wr, wqe_idx, max_sge);
852 		hr_qp->rq.wrid[wqe_idx] = wr->wr_id;
853 	}
854 
855 out:
856 	if (likely(nreq)) {
857 		hr_qp->rq.head += nreq;
858 
859 		update_rq_db(hr_dev, hr_qp);
860 	}
861 	spin_unlock_irqrestore(&hr_qp->rq.lock, flags);
862 
863 	return ret;
864 }
865 
866 static void *get_srq_wqe_buf(struct hns_roce_srq *srq, u32 n)
867 {
868 	return hns_roce_buf_offset(srq->buf_mtr.kmem, n << srq->wqe_shift);
869 }
870 
871 static void *get_idx_buf(struct hns_roce_idx_que *idx_que, u32 n)
872 {
873 	return hns_roce_buf_offset(idx_que->mtr.kmem,
874 				   n << idx_que->entry_shift);
875 }
876 
877 static void hns_roce_free_srq_wqe(struct hns_roce_srq *srq, u32 wqe_index)
878 {
879 	/* always called with interrupts disabled. */
880 	spin_lock(&srq->lock);
881 
882 	bitmap_clear(srq->idx_que.bitmap, wqe_index, 1);
883 	srq->idx_que.tail++;
884 
885 	spin_unlock(&srq->lock);
886 }
887 
888 static int hns_roce_srqwq_overflow(struct hns_roce_srq *srq)
889 {
890 	struct hns_roce_idx_que *idx_que = &srq->idx_que;
891 
892 	return idx_que->head - idx_que->tail >= srq->wqe_cnt;
893 }
894 
895 static int check_post_srq_valid(struct hns_roce_srq *srq, u32 max_sge,
896 				const struct ib_recv_wr *wr)
897 {
898 	struct ib_device *ib_dev = srq->ibsrq.device;
899 
900 	if (unlikely(wr->num_sge > max_sge)) {
901 		ibdev_err(ib_dev,
902 			  "failed to check sge, wr->num_sge = %d, max_sge = %u.\n",
903 			  wr->num_sge, max_sge);
904 		return -EINVAL;
905 	}
906 
907 	if (unlikely(hns_roce_srqwq_overflow(srq))) {
908 		ibdev_err(ib_dev,
909 			  "failed to check srqwq status, srqwq is full.\n");
910 		return -ENOMEM;
911 	}
912 
913 	return 0;
914 }
915 
916 static int get_srq_wqe_idx(struct hns_roce_srq *srq, u32 *wqe_idx)
917 {
918 	struct hns_roce_idx_que *idx_que = &srq->idx_que;
919 	u32 pos;
920 
921 	pos = find_first_zero_bit(idx_que->bitmap, srq->wqe_cnt);
922 	if (unlikely(pos == srq->wqe_cnt))
923 		return -ENOSPC;
924 
925 	bitmap_set(idx_que->bitmap, pos, 1);
926 	*wqe_idx = pos;
927 	return 0;
928 }
929 
930 static void fill_wqe_idx(struct hns_roce_srq *srq, unsigned int wqe_idx)
931 {
932 	struct hns_roce_idx_que *idx_que = &srq->idx_que;
933 	unsigned int head;
934 	__le32 *buf;
935 
936 	head = idx_que->head & (srq->wqe_cnt - 1);
937 
938 	buf = get_idx_buf(idx_que, head);
939 	*buf = cpu_to_le32(wqe_idx);
940 
941 	idx_que->head++;
942 }
943 
944 static void update_srq_db(struct hns_roce_srq *srq)
945 {
946 	struct hns_roce_dev *hr_dev = to_hr_dev(srq->ibsrq.device);
947 	struct hns_roce_v2_db db;
948 
949 	hr_reg_write(&db, DB_TAG, srq->srqn);
950 	hr_reg_write(&db, DB_CMD, HNS_ROCE_V2_SRQ_DB);
951 	hr_reg_write(&db, DB_PI, srq->idx_que.head);
952 
953 	hns_roce_write64(hr_dev, (__le32 *)&db, srq->db_reg);
954 }
955 
956 static int hns_roce_v2_post_srq_recv(struct ib_srq *ibsrq,
957 				     const struct ib_recv_wr *wr,
958 				     const struct ib_recv_wr **bad_wr)
959 {
960 	struct hns_roce_srq *srq = to_hr_srq(ibsrq);
961 	unsigned long flags;
962 	int ret = 0;
963 	u32 max_sge;
964 	u32 wqe_idx;
965 	void *wqe;
966 	u32 nreq;
967 
968 	spin_lock_irqsave(&srq->lock, flags);
969 
970 	max_sge = srq->max_gs - srq->rsv_sge;
971 	for (nreq = 0; wr; ++nreq, wr = wr->next) {
972 		ret = check_post_srq_valid(srq, max_sge, wr);
973 		if (ret) {
974 			*bad_wr = wr;
975 			break;
976 		}
977 
978 		ret = get_srq_wqe_idx(srq, &wqe_idx);
979 		if (unlikely(ret)) {
980 			*bad_wr = wr;
981 			break;
982 		}
983 
984 		wqe = get_srq_wqe_buf(srq, wqe_idx);
985 		fill_recv_sge_to_wqe(wr, wqe, max_sge, srq->rsv_sge);
986 		fill_wqe_idx(srq, wqe_idx);
987 		srq->wrid[wqe_idx] = wr->wr_id;
988 	}
989 
990 	if (likely(nreq)) {
991 		if (srq->cap_flags & HNS_ROCE_SRQ_CAP_RECORD_DB)
992 			*srq->rdb.db_record = srq->idx_que.head &
993 					      V2_DB_PRODUCER_IDX_M;
994 		else
995 			update_srq_db(srq);
996 	}
997 
998 	spin_unlock_irqrestore(&srq->lock, flags);
999 
1000 	return ret;
1001 }
1002 
1003 static u32 hns_roce_v2_cmd_hw_reseted(struct hns_roce_dev *hr_dev,
1004 				      unsigned long instance_stage,
1005 				      unsigned long reset_stage)
1006 {
1007 	/* When hardware reset has been completed once or more, we should stop
1008 	 * sending mailbox&cmq&doorbell to hardware. If now in .init_instance()
1009 	 * function, we should exit with error. If now at HNAE3_INIT_CLIENT
1010 	 * stage of soft reset process, we should exit with error, and then
1011 	 * HNAE3_INIT_CLIENT related process can rollback the operation like
1012 	 * notifing hardware to free resources, HNAE3_INIT_CLIENT related
1013 	 * process will exit with error to notify NIC driver to reschedule soft
1014 	 * reset process once again.
1015 	 */
1016 	hr_dev->is_reset = true;
1017 	hr_dev->dis_db = true;
1018 
1019 	if (reset_stage == HNS_ROCE_STATE_RST_INIT ||
1020 	    instance_stage == HNS_ROCE_STATE_INIT)
1021 		return CMD_RST_PRC_EBUSY;
1022 
1023 	return CMD_RST_PRC_SUCCESS;
1024 }
1025 
1026 static u32 hns_roce_v2_cmd_hw_resetting(struct hns_roce_dev *hr_dev,
1027 					unsigned long instance_stage,
1028 					unsigned long reset_stage)
1029 {
1030 #define HW_RESET_TIMEOUT_US 1000000
1031 #define HW_RESET_SLEEP_US 1000
1032 
1033 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1034 	struct hnae3_handle *handle = priv->handle;
1035 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1036 	unsigned long val;
1037 	int ret;
1038 
1039 	/* When hardware reset is detected, we should stop sending mailbox&cmq&
1040 	 * doorbell to hardware. If now in .init_instance() function, we should
1041 	 * exit with error. If now at HNAE3_INIT_CLIENT stage of soft reset
1042 	 * process, we should exit with error, and then HNAE3_INIT_CLIENT
1043 	 * related process can rollback the operation like notifing hardware to
1044 	 * free resources, HNAE3_INIT_CLIENT related process will exit with
1045 	 * error to notify NIC driver to reschedule soft reset process once
1046 	 * again.
1047 	 */
1048 	hr_dev->dis_db = true;
1049 
1050 	ret = read_poll_timeout(ops->ae_dev_reset_cnt, val,
1051 				val > hr_dev->reset_cnt, HW_RESET_SLEEP_US,
1052 				HW_RESET_TIMEOUT_US, false, handle);
1053 	if (!ret)
1054 		hr_dev->is_reset = true;
1055 
1056 	if (!hr_dev->is_reset || reset_stage == HNS_ROCE_STATE_RST_INIT ||
1057 	    instance_stage == HNS_ROCE_STATE_INIT)
1058 		return CMD_RST_PRC_EBUSY;
1059 
1060 	return CMD_RST_PRC_SUCCESS;
1061 }
1062 
1063 static u32 hns_roce_v2_cmd_sw_resetting(struct hns_roce_dev *hr_dev)
1064 {
1065 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1066 	struct hnae3_handle *handle = priv->handle;
1067 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1068 
1069 	/* When software reset is detected at .init_instance() function, we
1070 	 * should stop sending mailbox&cmq&doorbell to hardware, and exit
1071 	 * with error.
1072 	 */
1073 	hr_dev->dis_db = true;
1074 	if (ops->ae_dev_reset_cnt(handle) != hr_dev->reset_cnt)
1075 		hr_dev->is_reset = true;
1076 
1077 	return CMD_RST_PRC_EBUSY;
1078 }
1079 
1080 static u32 check_aedev_reset_status(struct hns_roce_dev *hr_dev,
1081 				    struct hnae3_handle *handle)
1082 {
1083 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1084 	unsigned long instance_stage; /* the current instance stage */
1085 	unsigned long reset_stage; /* the current reset stage */
1086 	unsigned long reset_cnt;
1087 	bool sw_resetting;
1088 	bool hw_resetting;
1089 
1090 	/* Get information about reset from NIC driver or RoCE driver itself,
1091 	 * the meaning of the following variables from NIC driver are described
1092 	 * as below:
1093 	 * reset_cnt -- The count value of completed hardware reset.
1094 	 * hw_resetting -- Whether hardware device is resetting now.
1095 	 * sw_resetting -- Whether NIC's software reset process is running now.
1096 	 */
1097 	instance_stage = handle->rinfo.instance_state;
1098 	reset_stage = handle->rinfo.reset_state;
1099 	reset_cnt = ops->ae_dev_reset_cnt(handle);
1100 	if (reset_cnt != hr_dev->reset_cnt)
1101 		return hns_roce_v2_cmd_hw_reseted(hr_dev, instance_stage,
1102 						  reset_stage);
1103 
1104 	hw_resetting = ops->get_cmdq_stat(handle);
1105 	if (hw_resetting)
1106 		return hns_roce_v2_cmd_hw_resetting(hr_dev, instance_stage,
1107 						    reset_stage);
1108 
1109 	sw_resetting = ops->ae_dev_resetting(handle);
1110 	if (sw_resetting && instance_stage == HNS_ROCE_STATE_INIT)
1111 		return hns_roce_v2_cmd_sw_resetting(hr_dev);
1112 
1113 	return CMD_RST_PRC_OTHERS;
1114 }
1115 
1116 static bool check_device_is_in_reset(struct hns_roce_dev *hr_dev)
1117 {
1118 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1119 	struct hnae3_handle *handle = priv->handle;
1120 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1121 
1122 	if (hr_dev->reset_cnt != ops->ae_dev_reset_cnt(handle))
1123 		return true;
1124 
1125 	if (ops->get_hw_reset_stat(handle))
1126 		return true;
1127 
1128 	if (ops->ae_dev_resetting(handle))
1129 		return true;
1130 
1131 	return false;
1132 }
1133 
1134 static bool v2_chk_mbox_is_avail(struct hns_roce_dev *hr_dev, bool *busy)
1135 {
1136 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1137 	u32 status;
1138 
1139 	if (hr_dev->is_reset)
1140 		status = CMD_RST_PRC_SUCCESS;
1141 	else
1142 		status = check_aedev_reset_status(hr_dev, priv->handle);
1143 
1144 	*busy = (status == CMD_RST_PRC_EBUSY);
1145 
1146 	return status == CMD_RST_PRC_OTHERS;
1147 }
1148 
1149 static int hns_roce_alloc_cmq_desc(struct hns_roce_dev *hr_dev,
1150 				   struct hns_roce_v2_cmq_ring *ring)
1151 {
1152 	int size = ring->desc_num * sizeof(struct hns_roce_cmq_desc);
1153 
1154 	ring->desc = dma_alloc_coherent(hr_dev->dev, size,
1155 					&ring->desc_dma_addr, GFP_KERNEL);
1156 	if (!ring->desc)
1157 		return -ENOMEM;
1158 
1159 	return 0;
1160 }
1161 
1162 static void hns_roce_free_cmq_desc(struct hns_roce_dev *hr_dev,
1163 				   struct hns_roce_v2_cmq_ring *ring)
1164 {
1165 	dma_free_coherent(hr_dev->dev,
1166 			  ring->desc_num * sizeof(struct hns_roce_cmq_desc),
1167 			  ring->desc, ring->desc_dma_addr);
1168 
1169 	ring->desc_dma_addr = 0;
1170 }
1171 
1172 static int init_csq(struct hns_roce_dev *hr_dev,
1173 		    struct hns_roce_v2_cmq_ring *csq)
1174 {
1175 	dma_addr_t dma;
1176 	int ret;
1177 
1178 	csq->desc_num = CMD_CSQ_DESC_NUM;
1179 	spin_lock_init(&csq->lock);
1180 	csq->flag = TYPE_CSQ;
1181 	csq->head = 0;
1182 
1183 	ret = hns_roce_alloc_cmq_desc(hr_dev, csq);
1184 	if (ret)
1185 		return ret;
1186 
1187 	dma = csq->desc_dma_addr;
1188 	roce_write(hr_dev, ROCEE_TX_CMQ_BASEADDR_L_REG, lower_32_bits(dma));
1189 	roce_write(hr_dev, ROCEE_TX_CMQ_BASEADDR_H_REG, upper_32_bits(dma));
1190 	roce_write(hr_dev, ROCEE_TX_CMQ_DEPTH_REG,
1191 		   (u32)csq->desc_num >> HNS_ROCE_CMQ_DESC_NUM_S);
1192 
1193 	/* Make sure to write CI first and then PI */
1194 	roce_write(hr_dev, ROCEE_TX_CMQ_CI_REG, 0);
1195 	roce_write(hr_dev, ROCEE_TX_CMQ_PI_REG, 0);
1196 
1197 	return 0;
1198 }
1199 
1200 static int hns_roce_v2_cmq_init(struct hns_roce_dev *hr_dev)
1201 {
1202 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1203 	int ret;
1204 
1205 	priv->cmq.tx_timeout = HNS_ROCE_CMQ_TX_TIMEOUT;
1206 
1207 	ret = init_csq(hr_dev, &priv->cmq.csq);
1208 	if (ret)
1209 		dev_err(hr_dev->dev, "failed to init CSQ, ret = %d.\n", ret);
1210 
1211 	return ret;
1212 }
1213 
1214 static void hns_roce_v2_cmq_exit(struct hns_roce_dev *hr_dev)
1215 {
1216 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1217 
1218 	hns_roce_free_cmq_desc(hr_dev, &priv->cmq.csq);
1219 }
1220 
1221 static void hns_roce_cmq_setup_basic_desc(struct hns_roce_cmq_desc *desc,
1222 					  enum hns_roce_opcode_type opcode,
1223 					  bool is_read)
1224 {
1225 	memset((void *)desc, 0, sizeof(struct hns_roce_cmq_desc));
1226 	desc->opcode = cpu_to_le16(opcode);
1227 	desc->flag = cpu_to_le16(HNS_ROCE_CMD_FLAG_IN);
1228 	if (is_read)
1229 		desc->flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_WR);
1230 	else
1231 		desc->flag &= cpu_to_le16(~HNS_ROCE_CMD_FLAG_WR);
1232 }
1233 
1234 static int hns_roce_cmq_csq_done(struct hns_roce_dev *hr_dev)
1235 {
1236 	u32 tail = roce_read(hr_dev, ROCEE_TX_CMQ_CI_REG);
1237 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1238 
1239 	return tail == priv->cmq.csq.head;
1240 }
1241 
1242 static void update_cmdq_status(struct hns_roce_dev *hr_dev)
1243 {
1244 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1245 	struct hnae3_handle *handle = priv->handle;
1246 
1247 	if (handle->rinfo.reset_state == HNS_ROCE_STATE_RST_INIT ||
1248 	    handle->rinfo.instance_state == HNS_ROCE_STATE_INIT)
1249 		hr_dev->cmd.state = HNS_ROCE_CMDQ_STATE_FATAL_ERR;
1250 }
1251 
1252 static int hns_roce_cmd_err_convert_errno(u16 desc_ret)
1253 {
1254 	struct hns_roce_cmd_errcode errcode_table[] = {
1255 		{CMD_EXEC_SUCCESS, 0},
1256 		{CMD_NO_AUTH, -EPERM},
1257 		{CMD_NOT_EXIST, -EOPNOTSUPP},
1258 		{CMD_CRQ_FULL, -EXFULL},
1259 		{CMD_NEXT_ERR, -ENOSR},
1260 		{CMD_NOT_EXEC, -ENOTBLK},
1261 		{CMD_PARA_ERR, -EINVAL},
1262 		{CMD_RESULT_ERR, -ERANGE},
1263 		{CMD_TIMEOUT, -ETIME},
1264 		{CMD_HILINK_ERR, -ENOLINK},
1265 		{CMD_INFO_ILLEGAL, -ENXIO},
1266 		{CMD_INVALID, -EBADR},
1267 	};
1268 	u16 i;
1269 
1270 	for (i = 0; i < ARRAY_SIZE(errcode_table); i++)
1271 		if (desc_ret == errcode_table[i].return_status)
1272 			return errcode_table[i].errno;
1273 	return -EIO;
1274 }
1275 
1276 static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
1277 			       struct hns_roce_cmq_desc *desc, int num)
1278 {
1279 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1280 	struct hns_roce_v2_cmq_ring *csq = &priv->cmq.csq;
1281 	u32 timeout = 0;
1282 	u16 desc_ret;
1283 	u32 tail;
1284 	int ret;
1285 	int i;
1286 
1287 	spin_lock_bh(&csq->lock);
1288 
1289 	tail = csq->head;
1290 
1291 	for (i = 0; i < num; i++) {
1292 		csq->desc[csq->head++] = desc[i];
1293 		if (csq->head == csq->desc_num)
1294 			csq->head = 0;
1295 	}
1296 
1297 	/* Write to hardware */
1298 	roce_write(hr_dev, ROCEE_TX_CMQ_PI_REG, csq->head);
1299 
1300 	atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_CMDS_CNT]);
1301 
1302 	do {
1303 		if (hns_roce_cmq_csq_done(hr_dev))
1304 			break;
1305 		udelay(1);
1306 	} while (++timeout < priv->cmq.tx_timeout);
1307 
1308 	if (hns_roce_cmq_csq_done(hr_dev)) {
1309 		ret = 0;
1310 		for (i = 0; i < num; i++) {
1311 			/* check the result of hardware write back */
1312 			desc[i] = csq->desc[tail++];
1313 			if (tail == csq->desc_num)
1314 				tail = 0;
1315 
1316 			desc_ret = le16_to_cpu(desc[i].retval);
1317 			if (likely(desc_ret == CMD_EXEC_SUCCESS))
1318 				continue;
1319 
1320 			dev_err_ratelimited(hr_dev->dev,
1321 					    "Cmdq IO error, opcode = 0x%x, return = 0x%x.\n",
1322 					    desc->opcode, desc_ret);
1323 			ret = hns_roce_cmd_err_convert_errno(desc_ret);
1324 		}
1325 	} else {
1326 		/* FW/HW reset or incorrect number of desc */
1327 		tail = roce_read(hr_dev, ROCEE_TX_CMQ_CI_REG);
1328 		dev_warn(hr_dev->dev, "CMDQ move tail from %u to %u.\n",
1329 			 csq->head, tail);
1330 		csq->head = tail;
1331 
1332 		update_cmdq_status(hr_dev);
1333 
1334 		ret = -EAGAIN;
1335 	}
1336 
1337 	spin_unlock_bh(&csq->lock);
1338 
1339 	if (ret)
1340 		atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_CMDS_ERR_CNT]);
1341 
1342 	return ret;
1343 }
1344 
1345 static int hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
1346 			     struct hns_roce_cmq_desc *desc, int num)
1347 {
1348 	bool busy;
1349 	int ret;
1350 
1351 	if (hr_dev->cmd.state == HNS_ROCE_CMDQ_STATE_FATAL_ERR)
1352 		return -EIO;
1353 
1354 	if (!v2_chk_mbox_is_avail(hr_dev, &busy))
1355 		return busy ? -EBUSY : 0;
1356 
1357 	ret = __hns_roce_cmq_send(hr_dev, desc, num);
1358 	if (ret) {
1359 		if (!v2_chk_mbox_is_avail(hr_dev, &busy))
1360 			return busy ? -EBUSY : 0;
1361 	}
1362 
1363 	return ret;
1364 }
1365 
1366 static int config_hem_ba_to_hw(struct hns_roce_dev *hr_dev,
1367 			       dma_addr_t base_addr, u8 cmd, unsigned long tag)
1368 {
1369 	struct hns_roce_cmd_mailbox *mbox;
1370 	int ret;
1371 
1372 	mbox = hns_roce_alloc_cmd_mailbox(hr_dev);
1373 	if (IS_ERR(mbox))
1374 		return PTR_ERR(mbox);
1375 
1376 	ret = hns_roce_cmd_mbox(hr_dev, base_addr, mbox->dma, cmd, tag);
1377 	hns_roce_free_cmd_mailbox(hr_dev, mbox);
1378 	return ret;
1379 }
1380 
1381 static int hns_roce_cmq_query_hw_info(struct hns_roce_dev *hr_dev)
1382 {
1383 	struct hns_roce_query_version *resp;
1384 	struct hns_roce_cmq_desc desc;
1385 	int ret;
1386 
1387 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_HW_VER, true);
1388 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1389 	if (ret)
1390 		return ret;
1391 
1392 	resp = (struct hns_roce_query_version *)desc.data;
1393 	hr_dev->hw_rev = le16_to_cpu(resp->rocee_hw_version);
1394 	hr_dev->vendor_id = hr_dev->pci_dev->vendor;
1395 
1396 	return 0;
1397 }
1398 
1399 static void func_clr_hw_resetting_state(struct hns_roce_dev *hr_dev,
1400 					struct hnae3_handle *handle)
1401 {
1402 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1403 	unsigned long end;
1404 
1405 	hr_dev->dis_db = true;
1406 
1407 	dev_warn(hr_dev->dev,
1408 		 "func clear is pending, device in resetting state.\n");
1409 	end = HNS_ROCE_V2_HW_RST_TIMEOUT;
1410 	while (end) {
1411 		if (!ops->get_hw_reset_stat(handle)) {
1412 			hr_dev->is_reset = true;
1413 			dev_info(hr_dev->dev,
1414 				 "func clear success after reset.\n");
1415 			return;
1416 		}
1417 		msleep(HNS_ROCE_V2_HW_RST_COMPLETION_WAIT);
1418 		end -= HNS_ROCE_V2_HW_RST_COMPLETION_WAIT;
1419 	}
1420 
1421 	dev_warn(hr_dev->dev, "func clear failed.\n");
1422 }
1423 
1424 static void func_clr_sw_resetting_state(struct hns_roce_dev *hr_dev,
1425 					struct hnae3_handle *handle)
1426 {
1427 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1428 	unsigned long end;
1429 
1430 	hr_dev->dis_db = true;
1431 
1432 	dev_warn(hr_dev->dev,
1433 		 "func clear is pending, device in resetting state.\n");
1434 	end = HNS_ROCE_V2_HW_RST_TIMEOUT;
1435 	while (end) {
1436 		if (ops->ae_dev_reset_cnt(handle) !=
1437 		    hr_dev->reset_cnt) {
1438 			hr_dev->is_reset = true;
1439 			dev_info(hr_dev->dev,
1440 				 "func clear success after sw reset\n");
1441 			return;
1442 		}
1443 		msleep(HNS_ROCE_V2_HW_RST_COMPLETION_WAIT);
1444 		end -= HNS_ROCE_V2_HW_RST_COMPLETION_WAIT;
1445 	}
1446 
1447 	dev_warn(hr_dev->dev, "func clear failed because of unfinished sw reset\n");
1448 }
1449 
1450 static void hns_roce_func_clr_rst_proc(struct hns_roce_dev *hr_dev, int retval,
1451 				       int flag)
1452 {
1453 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1454 	struct hnae3_handle *handle = priv->handle;
1455 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1456 
1457 	if (ops->ae_dev_reset_cnt(handle) != hr_dev->reset_cnt) {
1458 		hr_dev->dis_db = true;
1459 		hr_dev->is_reset = true;
1460 		dev_info(hr_dev->dev, "func clear success after reset.\n");
1461 		return;
1462 	}
1463 
1464 	if (ops->get_hw_reset_stat(handle)) {
1465 		func_clr_hw_resetting_state(hr_dev, handle);
1466 		return;
1467 	}
1468 
1469 	if (ops->ae_dev_resetting(handle) &&
1470 	    handle->rinfo.instance_state == HNS_ROCE_STATE_INIT) {
1471 		func_clr_sw_resetting_state(hr_dev, handle);
1472 		return;
1473 	}
1474 
1475 	if (retval && !flag)
1476 		dev_warn(hr_dev->dev,
1477 			 "func clear read failed, ret = %d.\n", retval);
1478 
1479 	dev_warn(hr_dev->dev, "func clear failed.\n");
1480 }
1481 
1482 static void __hns_roce_function_clear(struct hns_roce_dev *hr_dev, int vf_id)
1483 {
1484 	bool fclr_write_fail_flag = false;
1485 	struct hns_roce_func_clear *resp;
1486 	struct hns_roce_cmq_desc desc;
1487 	unsigned long end;
1488 	int ret = 0;
1489 
1490 	if (check_device_is_in_reset(hr_dev))
1491 		goto out;
1492 
1493 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_FUNC_CLEAR, false);
1494 	resp = (struct hns_roce_func_clear *)desc.data;
1495 	resp->rst_funcid_en = cpu_to_le32(vf_id);
1496 
1497 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1498 	if (ret) {
1499 		fclr_write_fail_flag = true;
1500 		dev_err(hr_dev->dev, "func clear write failed, ret = %d.\n",
1501 			 ret);
1502 		goto out;
1503 	}
1504 
1505 	msleep(HNS_ROCE_V2_READ_FUNC_CLEAR_FLAG_INTERVAL);
1506 	end = HNS_ROCE_V2_FUNC_CLEAR_TIMEOUT_MSECS;
1507 	while (end) {
1508 		if (check_device_is_in_reset(hr_dev))
1509 			goto out;
1510 		msleep(HNS_ROCE_V2_READ_FUNC_CLEAR_FLAG_FAIL_WAIT);
1511 		end -= HNS_ROCE_V2_READ_FUNC_CLEAR_FLAG_FAIL_WAIT;
1512 
1513 		hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_FUNC_CLEAR,
1514 					      true);
1515 
1516 		resp->rst_funcid_en = cpu_to_le32(vf_id);
1517 		ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1518 		if (ret)
1519 			continue;
1520 
1521 		if (hr_reg_read(resp, FUNC_CLEAR_RST_FUN_DONE)) {
1522 			if (vf_id == 0)
1523 				hr_dev->is_reset = true;
1524 			return;
1525 		}
1526 	}
1527 
1528 out:
1529 	hns_roce_func_clr_rst_proc(hr_dev, ret, fclr_write_fail_flag);
1530 }
1531 
1532 static int hns_roce_free_vf_resource(struct hns_roce_dev *hr_dev, int vf_id)
1533 {
1534 	enum hns_roce_opcode_type opcode = HNS_ROCE_OPC_ALLOC_VF_RES;
1535 	struct hns_roce_cmq_desc desc[2];
1536 	struct hns_roce_cmq_req *req_a;
1537 
1538 	req_a = (struct hns_roce_cmq_req *)desc[0].data;
1539 	hns_roce_cmq_setup_basic_desc(&desc[0], opcode, false);
1540 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
1541 	hns_roce_cmq_setup_basic_desc(&desc[1], opcode, false);
1542 	hr_reg_write(req_a, FUNC_RES_A_VF_ID, vf_id);
1543 
1544 	return hns_roce_cmq_send(hr_dev, desc, 2);
1545 }
1546 
1547 static void hns_roce_function_clear(struct hns_roce_dev *hr_dev)
1548 {
1549 	int ret;
1550 	int i;
1551 
1552 	if (hr_dev->cmd.state == HNS_ROCE_CMDQ_STATE_FATAL_ERR)
1553 		return;
1554 
1555 	for (i = hr_dev->func_num - 1; i >= 0; i--) {
1556 		__hns_roce_function_clear(hr_dev, i);
1557 
1558 		if (i == 0)
1559 			continue;
1560 
1561 		ret = hns_roce_free_vf_resource(hr_dev, i);
1562 		if (ret)
1563 			ibdev_err(&hr_dev->ib_dev,
1564 				  "failed to free vf resource, vf_id = %d, ret = %d.\n",
1565 				  i, ret);
1566 	}
1567 }
1568 
1569 static int hns_roce_clear_extdb_list_info(struct hns_roce_dev *hr_dev)
1570 {
1571 	struct hns_roce_cmq_desc desc;
1572 	int ret;
1573 
1574 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CLEAR_EXTDB_LIST_INFO,
1575 				      false);
1576 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1577 	if (ret)
1578 		ibdev_err(&hr_dev->ib_dev,
1579 			  "failed to clear extended doorbell info, ret = %d.\n",
1580 			  ret);
1581 
1582 	return ret;
1583 }
1584 
1585 static int hns_roce_query_fw_ver(struct hns_roce_dev *hr_dev)
1586 {
1587 	struct hns_roce_query_fw_info *resp;
1588 	struct hns_roce_cmq_desc desc;
1589 	int ret;
1590 
1591 	hns_roce_cmq_setup_basic_desc(&desc, HNS_QUERY_FW_VER, true);
1592 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1593 	if (ret)
1594 		return ret;
1595 
1596 	resp = (struct hns_roce_query_fw_info *)desc.data;
1597 	hr_dev->caps.fw_ver = (u64)(le32_to_cpu(resp->fw_ver));
1598 
1599 	return 0;
1600 }
1601 
1602 static int hns_roce_query_func_info(struct hns_roce_dev *hr_dev)
1603 {
1604 	struct hns_roce_cmq_desc desc;
1605 	int ret;
1606 
1607 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
1608 		hr_dev->func_num = 1;
1609 		return 0;
1610 	}
1611 
1612 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_FUNC_INFO,
1613 				      true);
1614 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1615 	if (ret) {
1616 		hr_dev->func_num = 1;
1617 		return ret;
1618 	}
1619 
1620 	hr_dev->func_num = le32_to_cpu(desc.func_info.own_func_num);
1621 	hr_dev->cong_algo_tmpl_id = le32_to_cpu(desc.func_info.own_mac_id);
1622 
1623 	return 0;
1624 }
1625 
1626 static int hns_roce_hw_v2_query_counter(struct hns_roce_dev *hr_dev,
1627 					u64 *stats, u32 port, int *num_counters)
1628 {
1629 #define CNT_PER_DESC 3
1630 	struct hns_roce_cmq_desc *desc;
1631 	int bd_idx, cnt_idx;
1632 	__le64 *cnt_data;
1633 	int desc_num;
1634 	int ret;
1635 	int i;
1636 
1637 	if (port > hr_dev->caps.num_ports)
1638 		return -EINVAL;
1639 
1640 	desc_num = DIV_ROUND_UP(HNS_ROCE_HW_CNT_TOTAL, CNT_PER_DESC);
1641 	desc = kcalloc(desc_num, sizeof(*desc), GFP_KERNEL);
1642 	if (!desc)
1643 		return -ENOMEM;
1644 
1645 	for (i = 0; i < desc_num; i++) {
1646 		hns_roce_cmq_setup_basic_desc(&desc[i],
1647 					      HNS_ROCE_OPC_QUERY_COUNTER, true);
1648 		if (i != desc_num - 1)
1649 			desc[i].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
1650 	}
1651 
1652 	ret = hns_roce_cmq_send(hr_dev, desc, desc_num);
1653 	if (ret) {
1654 		ibdev_err(&hr_dev->ib_dev,
1655 			  "failed to get counter, ret = %d.\n", ret);
1656 		goto err_out;
1657 	}
1658 
1659 	for (i = 0; i < HNS_ROCE_HW_CNT_TOTAL && i < *num_counters; i++) {
1660 		bd_idx = i / CNT_PER_DESC;
1661 		if (!(desc[bd_idx].flag & HNS_ROCE_CMD_FLAG_NEXT) &&
1662 		    bd_idx != HNS_ROCE_HW_CNT_TOTAL / CNT_PER_DESC)
1663 			break;
1664 
1665 		cnt_data = (__le64 *)&desc[bd_idx].data[0];
1666 		cnt_idx = i % CNT_PER_DESC;
1667 		stats[i] = le64_to_cpu(cnt_data[cnt_idx]);
1668 	}
1669 	*num_counters = i;
1670 
1671 err_out:
1672 	kfree(desc);
1673 	return ret;
1674 }
1675 
1676 static int hns_roce_config_global_param(struct hns_roce_dev *hr_dev)
1677 {
1678 	struct hns_roce_cmq_desc desc;
1679 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
1680 	u32 clock_cycles_of_1us;
1681 
1682 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_GLOBAL_PARAM,
1683 				      false);
1684 
1685 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08)
1686 		clock_cycles_of_1us = HNS_ROCE_1NS_CFG;
1687 	else
1688 		clock_cycles_of_1us = HNS_ROCE_1US_CFG;
1689 
1690 	hr_reg_write(req, CFG_GLOBAL_PARAM_1US_CYCLES, clock_cycles_of_1us);
1691 	hr_reg_write(req, CFG_GLOBAL_PARAM_UDP_PORT, ROCE_V2_UDP_DPORT);
1692 
1693 	return hns_roce_cmq_send(hr_dev, &desc, 1);
1694 }
1695 
1696 static int load_func_res_caps(struct hns_roce_dev *hr_dev, bool is_vf)
1697 {
1698 	struct hns_roce_cmq_desc desc[2];
1699 	struct hns_roce_cmq_req *r_a = (struct hns_roce_cmq_req *)desc[0].data;
1700 	struct hns_roce_cmq_req *r_b = (struct hns_roce_cmq_req *)desc[1].data;
1701 	struct hns_roce_caps *caps = &hr_dev->caps;
1702 	enum hns_roce_opcode_type opcode;
1703 	u32 func_num;
1704 	int ret;
1705 
1706 	if (is_vf) {
1707 		opcode = HNS_ROCE_OPC_QUERY_VF_RES;
1708 		func_num = 1;
1709 	} else {
1710 		opcode = HNS_ROCE_OPC_QUERY_PF_RES;
1711 		func_num = hr_dev->func_num;
1712 	}
1713 
1714 	hns_roce_cmq_setup_basic_desc(&desc[0], opcode, true);
1715 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
1716 	hns_roce_cmq_setup_basic_desc(&desc[1], opcode, true);
1717 
1718 	ret = hns_roce_cmq_send(hr_dev, desc, 2);
1719 	if (ret)
1720 		return ret;
1721 
1722 	caps->qpc_bt_num = hr_reg_read(r_a, FUNC_RES_A_QPC_BT_NUM) / func_num;
1723 	caps->srqc_bt_num = hr_reg_read(r_a, FUNC_RES_A_SRQC_BT_NUM) / func_num;
1724 	caps->cqc_bt_num = hr_reg_read(r_a, FUNC_RES_A_CQC_BT_NUM) / func_num;
1725 	caps->mpt_bt_num = hr_reg_read(r_a, FUNC_RES_A_MPT_BT_NUM) / func_num;
1726 	caps->eqc_bt_num = hr_reg_read(r_a, FUNC_RES_A_EQC_BT_NUM) / func_num;
1727 	caps->smac_bt_num = hr_reg_read(r_b, FUNC_RES_B_SMAC_NUM) / func_num;
1728 	caps->sgid_bt_num = hr_reg_read(r_b, FUNC_RES_B_SGID_NUM) / func_num;
1729 	caps->sccc_bt_num = hr_reg_read(r_b, FUNC_RES_B_SCCC_BT_NUM) / func_num;
1730 
1731 	if (is_vf) {
1732 		caps->sl_num = hr_reg_read(r_b, FUNC_RES_V_QID_NUM) / func_num;
1733 		caps->gmv_bt_num = hr_reg_read(r_b, FUNC_RES_V_GMV_BT_NUM) /
1734 					       func_num;
1735 	} else {
1736 		caps->sl_num = hr_reg_read(r_b, FUNC_RES_B_QID_NUM) / func_num;
1737 		caps->gmv_bt_num = hr_reg_read(r_b, FUNC_RES_B_GMV_BT_NUM) /
1738 					       func_num;
1739 	}
1740 
1741 	return 0;
1742 }
1743 
1744 static int load_pf_timer_res_caps(struct hns_roce_dev *hr_dev)
1745 {
1746 	struct hns_roce_cmq_desc desc;
1747 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
1748 	struct hns_roce_caps *caps = &hr_dev->caps;
1749 	int ret;
1750 
1751 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_PF_TIMER_RES,
1752 				      true);
1753 
1754 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1755 	if (ret)
1756 		return ret;
1757 
1758 	caps->qpc_timer_bt_num = hr_reg_read(req, PF_TIMER_RES_QPC_ITEM_NUM);
1759 	caps->cqc_timer_bt_num = hr_reg_read(req, PF_TIMER_RES_CQC_ITEM_NUM);
1760 
1761 	return 0;
1762 }
1763 
1764 static int hns_roce_query_pf_resource(struct hns_roce_dev *hr_dev)
1765 {
1766 	struct device *dev = hr_dev->dev;
1767 	int ret;
1768 
1769 	ret = load_func_res_caps(hr_dev, false);
1770 	if (ret) {
1771 		dev_err(dev, "failed to load pf res caps, ret = %d.\n", ret);
1772 		return ret;
1773 	}
1774 
1775 	ret = load_pf_timer_res_caps(hr_dev);
1776 	if (ret)
1777 		dev_err(dev, "failed to load pf timer resource, ret = %d.\n",
1778 			ret);
1779 
1780 	return ret;
1781 }
1782 
1783 static int hns_roce_query_vf_resource(struct hns_roce_dev *hr_dev)
1784 {
1785 	struct device *dev = hr_dev->dev;
1786 	int ret;
1787 
1788 	ret = load_func_res_caps(hr_dev, true);
1789 	if (ret)
1790 		dev_err(dev, "failed to load vf res caps, ret = %d.\n", ret);
1791 
1792 	return ret;
1793 }
1794 
1795 static int __hns_roce_set_vf_switch_param(struct hns_roce_dev *hr_dev,
1796 					  u32 vf_id)
1797 {
1798 	struct hns_roce_vf_switch *swt;
1799 	struct hns_roce_cmq_desc desc;
1800 	int ret;
1801 
1802 	swt = (struct hns_roce_vf_switch *)desc.data;
1803 	hns_roce_cmq_setup_basic_desc(&desc, HNS_SWITCH_PARAMETER_CFG, true);
1804 	swt->rocee_sel |= cpu_to_le32(HNS_ICL_SWITCH_CMD_ROCEE_SEL);
1805 	hr_reg_write(swt, VF_SWITCH_VF_ID, vf_id);
1806 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1807 	if (ret)
1808 		return ret;
1809 
1810 	desc.flag = cpu_to_le16(HNS_ROCE_CMD_FLAG_IN);
1811 	desc.flag &= cpu_to_le16(~HNS_ROCE_CMD_FLAG_WR);
1812 	hr_reg_enable(swt, VF_SWITCH_ALW_LPBK);
1813 	hr_reg_clear(swt, VF_SWITCH_ALW_LCL_LPBK);
1814 	hr_reg_enable(swt, VF_SWITCH_ALW_DST_OVRD);
1815 
1816 	return hns_roce_cmq_send(hr_dev, &desc, 1);
1817 }
1818 
1819 static int hns_roce_set_vf_switch_param(struct hns_roce_dev *hr_dev)
1820 {
1821 	u32 vf_id;
1822 	int ret;
1823 
1824 	for (vf_id = 0; vf_id < hr_dev->func_num; vf_id++) {
1825 		ret = __hns_roce_set_vf_switch_param(hr_dev, vf_id);
1826 		if (ret)
1827 			return ret;
1828 	}
1829 	return 0;
1830 }
1831 
1832 static int config_vf_hem_resource(struct hns_roce_dev *hr_dev, int vf_id)
1833 {
1834 	struct hns_roce_cmq_desc desc[2];
1835 	struct hns_roce_cmq_req *r_a = (struct hns_roce_cmq_req *)desc[0].data;
1836 	struct hns_roce_cmq_req *r_b = (struct hns_roce_cmq_req *)desc[1].data;
1837 	enum hns_roce_opcode_type opcode = HNS_ROCE_OPC_ALLOC_VF_RES;
1838 	struct hns_roce_caps *caps = &hr_dev->caps;
1839 
1840 	hns_roce_cmq_setup_basic_desc(&desc[0], opcode, false);
1841 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
1842 	hns_roce_cmq_setup_basic_desc(&desc[1], opcode, false);
1843 
1844 	hr_reg_write(r_a, FUNC_RES_A_VF_ID, vf_id);
1845 
1846 	hr_reg_write(r_a, FUNC_RES_A_QPC_BT_NUM, caps->qpc_bt_num);
1847 	hr_reg_write(r_a, FUNC_RES_A_QPC_BT_IDX, vf_id * caps->qpc_bt_num);
1848 	hr_reg_write(r_a, FUNC_RES_A_SRQC_BT_NUM, caps->srqc_bt_num);
1849 	hr_reg_write(r_a, FUNC_RES_A_SRQC_BT_IDX, vf_id * caps->srqc_bt_num);
1850 	hr_reg_write(r_a, FUNC_RES_A_CQC_BT_NUM, caps->cqc_bt_num);
1851 	hr_reg_write(r_a, FUNC_RES_A_CQC_BT_IDX, vf_id * caps->cqc_bt_num);
1852 	hr_reg_write(r_a, FUNC_RES_A_MPT_BT_NUM, caps->mpt_bt_num);
1853 	hr_reg_write(r_a, FUNC_RES_A_MPT_BT_IDX, vf_id * caps->mpt_bt_num);
1854 	hr_reg_write(r_a, FUNC_RES_A_EQC_BT_NUM, caps->eqc_bt_num);
1855 	hr_reg_write(r_a, FUNC_RES_A_EQC_BT_IDX, vf_id * caps->eqc_bt_num);
1856 	hr_reg_write(r_b, FUNC_RES_V_QID_NUM, caps->sl_num);
1857 	hr_reg_write(r_b, FUNC_RES_B_QID_IDX, vf_id * caps->sl_num);
1858 	hr_reg_write(r_b, FUNC_RES_B_SCCC_BT_NUM, caps->sccc_bt_num);
1859 	hr_reg_write(r_b, FUNC_RES_B_SCCC_BT_IDX, vf_id * caps->sccc_bt_num);
1860 
1861 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
1862 		hr_reg_write(r_b, FUNC_RES_V_GMV_BT_NUM, caps->gmv_bt_num);
1863 		hr_reg_write(r_b, FUNC_RES_B_GMV_BT_IDX,
1864 			     vf_id * caps->gmv_bt_num);
1865 	} else {
1866 		hr_reg_write(r_b, FUNC_RES_B_SGID_NUM, caps->sgid_bt_num);
1867 		hr_reg_write(r_b, FUNC_RES_B_SGID_IDX,
1868 			     vf_id * caps->sgid_bt_num);
1869 		hr_reg_write(r_b, FUNC_RES_B_SMAC_NUM, caps->smac_bt_num);
1870 		hr_reg_write(r_b, FUNC_RES_B_SMAC_IDX,
1871 			     vf_id * caps->smac_bt_num);
1872 	}
1873 
1874 	return hns_roce_cmq_send(hr_dev, desc, 2);
1875 }
1876 
1877 static int hns_roce_alloc_vf_resource(struct hns_roce_dev *hr_dev)
1878 {
1879 	u32 func_num = max_t(u32, 1, hr_dev->func_num);
1880 	u32 vf_id;
1881 	int ret;
1882 
1883 	for (vf_id = 0; vf_id < func_num; vf_id++) {
1884 		ret = config_vf_hem_resource(hr_dev, vf_id);
1885 		if (ret) {
1886 			dev_err(hr_dev->dev,
1887 				"failed to config vf-%u hem res, ret = %d.\n",
1888 				vf_id, ret);
1889 			return ret;
1890 		}
1891 	}
1892 
1893 	return 0;
1894 }
1895 
1896 static int hns_roce_v2_set_bt(struct hns_roce_dev *hr_dev)
1897 {
1898 	struct hns_roce_cmq_desc desc;
1899 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
1900 	struct hns_roce_caps *caps = &hr_dev->caps;
1901 
1902 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_BT_ATTR, false);
1903 
1904 	hr_reg_write(req, CFG_BT_ATTR_QPC_BA_PGSZ,
1905 		     caps->qpc_ba_pg_sz + PG_SHIFT_OFFSET);
1906 	hr_reg_write(req, CFG_BT_ATTR_QPC_BUF_PGSZ,
1907 		     caps->qpc_buf_pg_sz + PG_SHIFT_OFFSET);
1908 	hr_reg_write(req, CFG_BT_ATTR_QPC_HOPNUM,
1909 		     to_hr_hem_hopnum(caps->qpc_hop_num, caps->num_qps));
1910 
1911 	hr_reg_write(req, CFG_BT_ATTR_SRQC_BA_PGSZ,
1912 		     caps->srqc_ba_pg_sz + PG_SHIFT_OFFSET);
1913 	hr_reg_write(req, CFG_BT_ATTR_SRQC_BUF_PGSZ,
1914 		     caps->srqc_buf_pg_sz + PG_SHIFT_OFFSET);
1915 	hr_reg_write(req, CFG_BT_ATTR_SRQC_HOPNUM,
1916 		     to_hr_hem_hopnum(caps->srqc_hop_num, caps->num_srqs));
1917 
1918 	hr_reg_write(req, CFG_BT_ATTR_CQC_BA_PGSZ,
1919 		     caps->cqc_ba_pg_sz + PG_SHIFT_OFFSET);
1920 	hr_reg_write(req, CFG_BT_ATTR_CQC_BUF_PGSZ,
1921 		     caps->cqc_buf_pg_sz + PG_SHIFT_OFFSET);
1922 	hr_reg_write(req, CFG_BT_ATTR_CQC_HOPNUM,
1923 		     to_hr_hem_hopnum(caps->cqc_hop_num, caps->num_cqs));
1924 
1925 	hr_reg_write(req, CFG_BT_ATTR_MPT_BA_PGSZ,
1926 		     caps->mpt_ba_pg_sz + PG_SHIFT_OFFSET);
1927 	hr_reg_write(req, CFG_BT_ATTR_MPT_BUF_PGSZ,
1928 		     caps->mpt_buf_pg_sz + PG_SHIFT_OFFSET);
1929 	hr_reg_write(req, CFG_BT_ATTR_MPT_HOPNUM,
1930 		     to_hr_hem_hopnum(caps->mpt_hop_num, caps->num_mtpts));
1931 
1932 	hr_reg_write(req, CFG_BT_ATTR_SCCC_BA_PGSZ,
1933 		     caps->sccc_ba_pg_sz + PG_SHIFT_OFFSET);
1934 	hr_reg_write(req, CFG_BT_ATTR_SCCC_BUF_PGSZ,
1935 		     caps->sccc_buf_pg_sz + PG_SHIFT_OFFSET);
1936 	hr_reg_write(req, CFG_BT_ATTR_SCCC_HOPNUM,
1937 		     to_hr_hem_hopnum(caps->sccc_hop_num, caps->num_qps));
1938 
1939 	return hns_roce_cmq_send(hr_dev, &desc, 1);
1940 }
1941 
1942 static void calc_pg_sz(u32 obj_num, u32 obj_size, u32 hop_num, u32 ctx_bt_num,
1943 		       u32 *buf_page_size, u32 *bt_page_size, u32 hem_type)
1944 {
1945 	u64 obj_per_chunk;
1946 	u64 bt_chunk_size = PAGE_SIZE;
1947 	u64 buf_chunk_size = PAGE_SIZE;
1948 	u64 obj_per_chunk_default = buf_chunk_size / obj_size;
1949 
1950 	*buf_page_size = 0;
1951 	*bt_page_size = 0;
1952 
1953 	switch (hop_num) {
1954 	case 3:
1955 		obj_per_chunk = ctx_bt_num * (bt_chunk_size / BA_BYTE_LEN) *
1956 				(bt_chunk_size / BA_BYTE_LEN) *
1957 				(bt_chunk_size / BA_BYTE_LEN) *
1958 				 obj_per_chunk_default;
1959 		break;
1960 	case 2:
1961 		obj_per_chunk = ctx_bt_num * (bt_chunk_size / BA_BYTE_LEN) *
1962 				(bt_chunk_size / BA_BYTE_LEN) *
1963 				 obj_per_chunk_default;
1964 		break;
1965 	case 1:
1966 		obj_per_chunk = ctx_bt_num * (bt_chunk_size / BA_BYTE_LEN) *
1967 				obj_per_chunk_default;
1968 		break;
1969 	case HNS_ROCE_HOP_NUM_0:
1970 		obj_per_chunk = ctx_bt_num * obj_per_chunk_default;
1971 		break;
1972 	default:
1973 		pr_err("table %u not support hop_num = %u!\n", hem_type,
1974 		       hop_num);
1975 		return;
1976 	}
1977 
1978 	if (hem_type >= HEM_TYPE_MTT)
1979 		*bt_page_size = ilog2(DIV_ROUND_UP(obj_num, obj_per_chunk));
1980 	else
1981 		*buf_page_size = ilog2(DIV_ROUND_UP(obj_num, obj_per_chunk));
1982 }
1983 
1984 static void set_hem_page_size(struct hns_roce_dev *hr_dev)
1985 {
1986 	struct hns_roce_caps *caps = &hr_dev->caps;
1987 
1988 	/* EQ */
1989 	caps->eqe_ba_pg_sz = 0;
1990 	caps->eqe_buf_pg_sz = 0;
1991 
1992 	/* Link Table */
1993 	caps->llm_buf_pg_sz = 0;
1994 
1995 	/* MR */
1996 	caps->mpt_ba_pg_sz = 0;
1997 	caps->mpt_buf_pg_sz = 0;
1998 	caps->pbl_ba_pg_sz = HNS_ROCE_BA_PG_SZ_SUPPORTED_16K;
1999 	caps->pbl_buf_pg_sz = 0;
2000 	calc_pg_sz(caps->num_mtpts, caps->mtpt_entry_sz, caps->mpt_hop_num,
2001 		   caps->mpt_bt_num, &caps->mpt_buf_pg_sz, &caps->mpt_ba_pg_sz,
2002 		   HEM_TYPE_MTPT);
2003 
2004 	/* QP */
2005 	caps->qpc_ba_pg_sz = 0;
2006 	caps->qpc_buf_pg_sz = 0;
2007 	caps->qpc_timer_ba_pg_sz = 0;
2008 	caps->qpc_timer_buf_pg_sz = 0;
2009 	caps->sccc_ba_pg_sz = 0;
2010 	caps->sccc_buf_pg_sz = 0;
2011 	caps->mtt_ba_pg_sz = 0;
2012 	caps->mtt_buf_pg_sz = 0;
2013 	calc_pg_sz(caps->num_qps, caps->qpc_sz, caps->qpc_hop_num,
2014 		   caps->qpc_bt_num, &caps->qpc_buf_pg_sz, &caps->qpc_ba_pg_sz,
2015 		   HEM_TYPE_QPC);
2016 
2017 	if (caps->flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL)
2018 		calc_pg_sz(caps->num_qps, caps->sccc_sz, caps->sccc_hop_num,
2019 			   caps->sccc_bt_num, &caps->sccc_buf_pg_sz,
2020 			   &caps->sccc_ba_pg_sz, HEM_TYPE_SCCC);
2021 
2022 	/* CQ */
2023 	caps->cqc_ba_pg_sz = 0;
2024 	caps->cqc_buf_pg_sz = 0;
2025 	caps->cqc_timer_ba_pg_sz = 0;
2026 	caps->cqc_timer_buf_pg_sz = 0;
2027 	caps->cqe_ba_pg_sz = HNS_ROCE_BA_PG_SZ_SUPPORTED_256K;
2028 	caps->cqe_buf_pg_sz = 0;
2029 	calc_pg_sz(caps->num_cqs, caps->cqc_entry_sz, caps->cqc_hop_num,
2030 		   caps->cqc_bt_num, &caps->cqc_buf_pg_sz, &caps->cqc_ba_pg_sz,
2031 		   HEM_TYPE_CQC);
2032 	calc_pg_sz(caps->max_cqes, caps->cqe_sz, caps->cqe_hop_num,
2033 		   1, &caps->cqe_buf_pg_sz, &caps->cqe_ba_pg_sz, HEM_TYPE_CQE);
2034 
2035 	/* SRQ */
2036 	if (caps->flags & HNS_ROCE_CAP_FLAG_SRQ) {
2037 		caps->srqc_ba_pg_sz = 0;
2038 		caps->srqc_buf_pg_sz = 0;
2039 		caps->srqwqe_ba_pg_sz = 0;
2040 		caps->srqwqe_buf_pg_sz = 0;
2041 		caps->idx_ba_pg_sz = 0;
2042 		caps->idx_buf_pg_sz = 0;
2043 		calc_pg_sz(caps->num_srqs, caps->srqc_entry_sz,
2044 			   caps->srqc_hop_num, caps->srqc_bt_num,
2045 			   &caps->srqc_buf_pg_sz, &caps->srqc_ba_pg_sz,
2046 			   HEM_TYPE_SRQC);
2047 		calc_pg_sz(caps->num_srqwqe_segs, caps->mtt_entry_sz,
2048 			   caps->srqwqe_hop_num, 1, &caps->srqwqe_buf_pg_sz,
2049 			   &caps->srqwqe_ba_pg_sz, HEM_TYPE_SRQWQE);
2050 		calc_pg_sz(caps->num_idx_segs, caps->idx_entry_sz,
2051 			   caps->idx_hop_num, 1, &caps->idx_buf_pg_sz,
2052 			   &caps->idx_ba_pg_sz, HEM_TYPE_IDX);
2053 	}
2054 
2055 	/* GMV */
2056 	caps->gmv_ba_pg_sz = 0;
2057 	caps->gmv_buf_pg_sz = 0;
2058 }
2059 
2060 /* Apply all loaded caps before setting to hardware */
2061 static void apply_func_caps(struct hns_roce_dev *hr_dev)
2062 {
2063 #define MAX_GID_TBL_LEN 256
2064 	struct hns_roce_caps *caps = &hr_dev->caps;
2065 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2066 
2067 	/* The following configurations don't need to be got from firmware. */
2068 	caps->qpc_timer_entry_sz = HNS_ROCE_V2_QPC_TIMER_ENTRY_SZ;
2069 	caps->cqc_timer_entry_sz = HNS_ROCE_V2_CQC_TIMER_ENTRY_SZ;
2070 	caps->mtt_entry_sz = HNS_ROCE_V2_MTT_ENTRY_SZ;
2071 
2072 	caps->pbl_hop_num = HNS_ROCE_PBL_HOP_NUM;
2073 	caps->qpc_timer_hop_num = HNS_ROCE_HOP_NUM_0;
2074 	caps->cqc_timer_hop_num = HNS_ROCE_HOP_NUM_0;
2075 
2076 	caps->num_srqwqe_segs = HNS_ROCE_V2_MAX_SRQWQE_SEGS;
2077 	caps->num_idx_segs = HNS_ROCE_V2_MAX_IDX_SEGS;
2078 
2079 	if (!caps->num_comp_vectors)
2080 		caps->num_comp_vectors =
2081 			min_t(u32, caps->eqc_bt_num - HNS_ROCE_V2_AEQE_VEC_NUM,
2082 				(u32)priv->handle->rinfo.num_vectors -
2083 		(HNS_ROCE_V2_AEQE_VEC_NUM + HNS_ROCE_V2_ABNORMAL_VEC_NUM));
2084 
2085 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
2086 		caps->eqe_hop_num = HNS_ROCE_V3_EQE_HOP_NUM;
2087 		caps->ceqe_size = HNS_ROCE_V3_EQE_SIZE;
2088 		caps->aeqe_size = HNS_ROCE_V3_EQE_SIZE;
2089 
2090 		/* The following configurations will be overwritten */
2091 		caps->qpc_sz = HNS_ROCE_V3_QPC_SZ;
2092 		caps->cqe_sz = HNS_ROCE_V3_CQE_SIZE;
2093 		caps->sccc_sz = HNS_ROCE_V3_SCCC_SZ;
2094 
2095 		/* The following configurations are not got from firmware */
2096 		caps->gmv_entry_sz = HNS_ROCE_V3_GMV_ENTRY_SZ;
2097 
2098 		caps->gmv_hop_num = HNS_ROCE_HOP_NUM_0;
2099 
2100 		/* It's meaningless to support excessively large gid_table_len,
2101 		 * as the type of sgid_index in kernel struct ib_global_route
2102 		 * and userspace struct ibv_global_route are u8/uint8_t (0-255).
2103 		 */
2104 		caps->gid_table_len[0] = min_t(u32, MAX_GID_TBL_LEN,
2105 					 caps->gmv_bt_num *
2106 					 (HNS_HW_PAGE_SIZE / caps->gmv_entry_sz));
2107 
2108 		caps->gmv_entry_num = caps->gmv_bt_num * (PAGE_SIZE /
2109 							  caps->gmv_entry_sz);
2110 	} else {
2111 		u32 func_num = max_t(u32, 1, hr_dev->func_num);
2112 
2113 		caps->eqe_hop_num = HNS_ROCE_V2_EQE_HOP_NUM;
2114 		caps->ceqe_size = HNS_ROCE_CEQE_SIZE;
2115 		caps->aeqe_size = HNS_ROCE_AEQE_SIZE;
2116 		caps->gid_table_len[0] /= func_num;
2117 	}
2118 
2119 	if (hr_dev->is_vf) {
2120 		caps->default_aeq_arm_st = 0x3;
2121 		caps->default_ceq_arm_st = 0x3;
2122 		caps->default_ceq_max_cnt = 0x1;
2123 		caps->default_ceq_period = 0x10;
2124 		caps->default_aeq_max_cnt = 0x1;
2125 		caps->default_aeq_period = 0x10;
2126 	}
2127 
2128 	set_hem_page_size(hr_dev);
2129 }
2130 
2131 static int hns_roce_query_caps(struct hns_roce_dev *hr_dev)
2132 {
2133 	struct hns_roce_cmq_desc desc[HNS_ROCE_QUERY_PF_CAPS_CMD_NUM];
2134 	struct hns_roce_caps *caps = &hr_dev->caps;
2135 	struct hns_roce_query_pf_caps_a *resp_a;
2136 	struct hns_roce_query_pf_caps_b *resp_b;
2137 	struct hns_roce_query_pf_caps_c *resp_c;
2138 	struct hns_roce_query_pf_caps_d *resp_d;
2139 	struct hns_roce_query_pf_caps_e *resp_e;
2140 	enum hns_roce_opcode_type cmd;
2141 	int ctx_hop_num;
2142 	int pbl_hop_num;
2143 	int ret;
2144 	int i;
2145 
2146 	cmd = hr_dev->is_vf ? HNS_ROCE_OPC_QUERY_VF_CAPS_NUM :
2147 	      HNS_ROCE_OPC_QUERY_PF_CAPS_NUM;
2148 
2149 	for (i = 0; i < HNS_ROCE_QUERY_PF_CAPS_CMD_NUM; i++) {
2150 		hns_roce_cmq_setup_basic_desc(&desc[i], cmd, true);
2151 		if (i < (HNS_ROCE_QUERY_PF_CAPS_CMD_NUM - 1))
2152 			desc[i].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
2153 		else
2154 			desc[i].flag &= ~cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
2155 	}
2156 
2157 	ret = hns_roce_cmq_send(hr_dev, desc, HNS_ROCE_QUERY_PF_CAPS_CMD_NUM);
2158 	if (ret)
2159 		return ret;
2160 
2161 	resp_a = (struct hns_roce_query_pf_caps_a *)desc[0].data;
2162 	resp_b = (struct hns_roce_query_pf_caps_b *)desc[1].data;
2163 	resp_c = (struct hns_roce_query_pf_caps_c *)desc[2].data;
2164 	resp_d = (struct hns_roce_query_pf_caps_d *)desc[3].data;
2165 	resp_e = (struct hns_roce_query_pf_caps_e *)desc[4].data;
2166 
2167 	caps->local_ca_ack_delay = resp_a->local_ca_ack_delay;
2168 	caps->max_sq_sg = le16_to_cpu(resp_a->max_sq_sg);
2169 	caps->max_sq_inline = le16_to_cpu(resp_a->max_sq_inline);
2170 	caps->max_rq_sg = le16_to_cpu(resp_a->max_rq_sg);
2171 	caps->max_rq_sg = roundup_pow_of_two(caps->max_rq_sg);
2172 	caps->max_srq_sges = le16_to_cpu(resp_a->max_srq_sges);
2173 	caps->max_srq_sges = roundup_pow_of_two(caps->max_srq_sges);
2174 	caps->num_aeq_vectors = resp_a->num_aeq_vectors;
2175 	caps->num_other_vectors = resp_a->num_other_vectors;
2176 	caps->max_sq_desc_sz = resp_a->max_sq_desc_sz;
2177 	caps->max_rq_desc_sz = resp_a->max_rq_desc_sz;
2178 
2179 	caps->mtpt_entry_sz = resp_b->mtpt_entry_sz;
2180 	caps->irrl_entry_sz = resp_b->irrl_entry_sz;
2181 	caps->trrl_entry_sz = resp_b->trrl_entry_sz;
2182 	caps->cqc_entry_sz = resp_b->cqc_entry_sz;
2183 	caps->srqc_entry_sz = resp_b->srqc_entry_sz;
2184 	caps->idx_entry_sz = resp_b->idx_entry_sz;
2185 	caps->sccc_sz = resp_b->sccc_sz;
2186 	caps->max_mtu = resp_b->max_mtu;
2187 	caps->min_cqes = resp_b->min_cqes;
2188 	caps->min_wqes = resp_b->min_wqes;
2189 	caps->page_size_cap = le32_to_cpu(resp_b->page_size_cap);
2190 	caps->pkey_table_len[0] = resp_b->pkey_table_len;
2191 	caps->phy_num_uars = resp_b->phy_num_uars;
2192 	ctx_hop_num = resp_b->ctx_hop_num;
2193 	pbl_hop_num = resp_b->pbl_hop_num;
2194 
2195 	caps->num_pds = 1 << hr_reg_read(resp_c, PF_CAPS_C_NUM_PDS);
2196 
2197 	caps->flags = hr_reg_read(resp_c, PF_CAPS_C_CAP_FLAGS);
2198 	caps->flags |= le16_to_cpu(resp_d->cap_flags_ex) <<
2199 		       HNS_ROCE_CAP_FLAGS_EX_SHIFT;
2200 
2201 	caps->num_cqs = 1 << hr_reg_read(resp_c, PF_CAPS_C_NUM_CQS);
2202 	caps->gid_table_len[0] = hr_reg_read(resp_c, PF_CAPS_C_MAX_GID);
2203 	caps->max_cqes = 1 << hr_reg_read(resp_c, PF_CAPS_C_CQ_DEPTH);
2204 	caps->num_xrcds = 1 << hr_reg_read(resp_c, PF_CAPS_C_NUM_XRCDS);
2205 	caps->num_mtpts = 1 << hr_reg_read(resp_c, PF_CAPS_C_NUM_MRWS);
2206 	caps->num_qps = 1 << hr_reg_read(resp_c, PF_CAPS_C_NUM_QPS);
2207 	caps->max_qp_init_rdma = hr_reg_read(resp_c, PF_CAPS_C_MAX_ORD);
2208 	caps->max_qp_dest_rdma = caps->max_qp_init_rdma;
2209 	caps->max_wqes = 1 << le16_to_cpu(resp_c->sq_depth);
2210 
2211 	caps->num_srqs = 1 << hr_reg_read(resp_d, PF_CAPS_D_NUM_SRQS);
2212 	caps->cong_type = hr_reg_read(resp_d, PF_CAPS_D_CONG_TYPE);
2213 	caps->max_srq_wrs = 1 << le16_to_cpu(resp_d->srq_depth);
2214 	caps->ceqe_depth = 1 << hr_reg_read(resp_d, PF_CAPS_D_CEQ_DEPTH);
2215 	caps->num_comp_vectors = hr_reg_read(resp_d, PF_CAPS_D_NUM_CEQS);
2216 	caps->aeqe_depth = 1 << hr_reg_read(resp_d, PF_CAPS_D_AEQ_DEPTH);
2217 	caps->reserved_pds = hr_reg_read(resp_d, PF_CAPS_D_RSV_PDS);
2218 	caps->num_uars = 1 << hr_reg_read(resp_d, PF_CAPS_D_NUM_UARS);
2219 	caps->reserved_qps = hr_reg_read(resp_d, PF_CAPS_D_RSV_QPS);
2220 	caps->reserved_uars = hr_reg_read(resp_d, PF_CAPS_D_RSV_UARS);
2221 
2222 	caps->reserved_mrws = hr_reg_read(resp_e, PF_CAPS_E_RSV_MRWS);
2223 	caps->chunk_sz = 1 << hr_reg_read(resp_e, PF_CAPS_E_CHUNK_SIZE_SHIFT);
2224 	caps->reserved_cqs = hr_reg_read(resp_e, PF_CAPS_E_RSV_CQS);
2225 	caps->reserved_xrcds = hr_reg_read(resp_e, PF_CAPS_E_RSV_XRCDS);
2226 	caps->reserved_srqs = hr_reg_read(resp_e, PF_CAPS_E_RSV_SRQS);
2227 	caps->reserved_lkey = hr_reg_read(resp_e, PF_CAPS_E_RSV_LKEYS);
2228 
2229 	caps->qpc_hop_num = ctx_hop_num;
2230 	caps->sccc_hop_num = ctx_hop_num;
2231 	caps->srqc_hop_num = ctx_hop_num;
2232 	caps->cqc_hop_num = ctx_hop_num;
2233 	caps->mpt_hop_num = ctx_hop_num;
2234 	caps->mtt_hop_num = pbl_hop_num;
2235 	caps->cqe_hop_num = pbl_hop_num;
2236 	caps->srqwqe_hop_num = pbl_hop_num;
2237 	caps->idx_hop_num = pbl_hop_num;
2238 	caps->wqe_sq_hop_num = hr_reg_read(resp_d, PF_CAPS_D_SQWQE_HOP_NUM);
2239 	caps->wqe_sge_hop_num = hr_reg_read(resp_d, PF_CAPS_D_EX_SGE_HOP_NUM);
2240 	caps->wqe_rq_hop_num = hr_reg_read(resp_d, PF_CAPS_D_RQWQE_HOP_NUM);
2241 
2242 	if (!(caps->page_size_cap & PAGE_SIZE))
2243 		caps->page_size_cap = HNS_ROCE_V2_PAGE_SIZE_SUPPORTED;
2244 
2245 	if (!hr_dev->is_vf) {
2246 		caps->cqe_sz = resp_a->cqe_sz;
2247 		caps->qpc_sz = le16_to_cpu(resp_b->qpc_sz);
2248 		caps->default_aeq_arm_st =
2249 				hr_reg_read(resp_d, PF_CAPS_D_AEQ_ARM_ST);
2250 		caps->default_ceq_arm_st =
2251 				hr_reg_read(resp_d, PF_CAPS_D_CEQ_ARM_ST);
2252 		caps->default_ceq_max_cnt = le16_to_cpu(resp_e->ceq_max_cnt);
2253 		caps->default_ceq_period = le16_to_cpu(resp_e->ceq_period);
2254 		caps->default_aeq_max_cnt = le16_to_cpu(resp_e->aeq_max_cnt);
2255 		caps->default_aeq_period = le16_to_cpu(resp_e->aeq_period);
2256 	}
2257 
2258 	return 0;
2259 }
2260 
2261 static int config_hem_entry_size(struct hns_roce_dev *hr_dev, u32 type, u32 val)
2262 {
2263 	struct hns_roce_cmq_desc desc;
2264 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
2265 
2266 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_ENTRY_SIZE,
2267 				      false);
2268 
2269 	hr_reg_write(req, CFG_HEM_ENTRY_SIZE_TYPE, type);
2270 	hr_reg_write(req, CFG_HEM_ENTRY_SIZE_VALUE, val);
2271 
2272 	return hns_roce_cmq_send(hr_dev, &desc, 1);
2273 }
2274 
2275 static int hns_roce_config_entry_size(struct hns_roce_dev *hr_dev)
2276 {
2277 	struct hns_roce_caps *caps = &hr_dev->caps;
2278 	int ret;
2279 
2280 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08)
2281 		return 0;
2282 
2283 	ret = config_hem_entry_size(hr_dev, HNS_ROCE_CFG_QPC_SIZE,
2284 				    caps->qpc_sz);
2285 	if (ret) {
2286 		dev_err(hr_dev->dev, "failed to cfg qpc sz, ret = %d.\n", ret);
2287 		return ret;
2288 	}
2289 
2290 	ret = config_hem_entry_size(hr_dev, HNS_ROCE_CFG_SCCC_SIZE,
2291 				    caps->sccc_sz);
2292 	if (ret)
2293 		dev_err(hr_dev->dev, "failed to cfg sccc sz, ret = %d.\n", ret);
2294 
2295 	return ret;
2296 }
2297 
2298 static int hns_roce_v2_vf_profile(struct hns_roce_dev *hr_dev)
2299 {
2300 	struct device *dev = hr_dev->dev;
2301 	int ret;
2302 
2303 	hr_dev->func_num = 1;
2304 
2305 	ret = hns_roce_query_caps(hr_dev);
2306 	if (ret) {
2307 		dev_err(dev, "failed to query VF caps, ret = %d.\n", ret);
2308 		return ret;
2309 	}
2310 
2311 	ret = hns_roce_query_vf_resource(hr_dev);
2312 	if (ret) {
2313 		dev_err(dev, "failed to query VF resource, ret = %d.\n", ret);
2314 		return ret;
2315 	}
2316 
2317 	apply_func_caps(hr_dev);
2318 
2319 	ret = hns_roce_v2_set_bt(hr_dev);
2320 	if (ret)
2321 		dev_err(dev, "failed to config VF BA table, ret = %d.\n", ret);
2322 
2323 	return ret;
2324 }
2325 
2326 static int hns_roce_v2_pf_profile(struct hns_roce_dev *hr_dev)
2327 {
2328 	struct device *dev = hr_dev->dev;
2329 	int ret;
2330 
2331 	ret = hns_roce_query_func_info(hr_dev);
2332 	if (ret) {
2333 		dev_err(dev, "failed to query func info, ret = %d.\n", ret);
2334 		return ret;
2335 	}
2336 
2337 	ret = hns_roce_config_global_param(hr_dev);
2338 	if (ret) {
2339 		dev_err(dev, "failed to config global param, ret = %d.\n", ret);
2340 		return ret;
2341 	}
2342 
2343 	ret = hns_roce_set_vf_switch_param(hr_dev);
2344 	if (ret) {
2345 		dev_err(dev, "failed to set switch param, ret = %d.\n", ret);
2346 		return ret;
2347 	}
2348 
2349 	ret = hns_roce_query_caps(hr_dev);
2350 	if (ret) {
2351 		dev_err(dev, "failed to query PF caps, ret = %d.\n", ret);
2352 		return ret;
2353 	}
2354 
2355 	ret = hns_roce_query_pf_resource(hr_dev);
2356 	if (ret) {
2357 		dev_err(dev, "failed to query pf resource, ret = %d.\n", ret);
2358 		return ret;
2359 	}
2360 
2361 	apply_func_caps(hr_dev);
2362 
2363 	ret = hns_roce_alloc_vf_resource(hr_dev);
2364 	if (ret) {
2365 		dev_err(dev, "failed to alloc vf resource, ret = %d.\n", ret);
2366 		return ret;
2367 	}
2368 
2369 	ret = hns_roce_v2_set_bt(hr_dev);
2370 	if (ret) {
2371 		dev_err(dev, "failed to config BA table, ret = %d.\n", ret);
2372 		return ret;
2373 	}
2374 
2375 	/* Configure the size of QPC, SCCC, etc. */
2376 	return hns_roce_config_entry_size(hr_dev);
2377 }
2378 
2379 static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev)
2380 {
2381 	struct device *dev = hr_dev->dev;
2382 	int ret;
2383 
2384 	ret = hns_roce_cmq_query_hw_info(hr_dev);
2385 	if (ret) {
2386 		dev_err(dev, "failed to query hardware info, ret = %d.\n", ret);
2387 		return ret;
2388 	}
2389 
2390 	ret = hns_roce_query_fw_ver(hr_dev);
2391 	if (ret) {
2392 		dev_err(dev, "failed to query firmware info, ret = %d.\n", ret);
2393 		return ret;
2394 	}
2395 
2396 	hr_dev->vendor_part_id = hr_dev->pci_dev->device;
2397 	hr_dev->sys_image_guid = be64_to_cpu(hr_dev->ib_dev.node_guid);
2398 
2399 	if (hr_dev->is_vf)
2400 		return hns_roce_v2_vf_profile(hr_dev);
2401 	else
2402 		return hns_roce_v2_pf_profile(hr_dev);
2403 }
2404 
2405 static void config_llm_table(struct hns_roce_buf *data_buf, void *cfg_buf)
2406 {
2407 	u32 i, next_ptr, page_num;
2408 	__le64 *entry = cfg_buf;
2409 	dma_addr_t addr;
2410 	u64 val;
2411 
2412 	page_num = data_buf->npages;
2413 	for (i = 0; i < page_num; i++) {
2414 		addr = hns_roce_buf_page(data_buf, i);
2415 		if (i == (page_num - 1))
2416 			next_ptr = 0;
2417 		else
2418 			next_ptr = i + 1;
2419 
2420 		val = HNS_ROCE_EXT_LLM_ENTRY(addr, (u64)next_ptr);
2421 		entry[i] = cpu_to_le64(val);
2422 	}
2423 }
2424 
2425 static int set_llm_cfg_to_hw(struct hns_roce_dev *hr_dev,
2426 			     struct hns_roce_link_table *table)
2427 {
2428 	struct hns_roce_cmq_desc desc[2];
2429 	struct hns_roce_cmq_req *r_a = (struct hns_roce_cmq_req *)desc[0].data;
2430 	struct hns_roce_cmq_req *r_b = (struct hns_roce_cmq_req *)desc[1].data;
2431 	struct hns_roce_buf *buf = table->buf;
2432 	enum hns_roce_opcode_type opcode;
2433 	dma_addr_t addr;
2434 
2435 	opcode = HNS_ROCE_OPC_CFG_EXT_LLM;
2436 	hns_roce_cmq_setup_basic_desc(&desc[0], opcode, false);
2437 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
2438 	hns_roce_cmq_setup_basic_desc(&desc[1], opcode, false);
2439 
2440 	hr_reg_write(r_a, CFG_LLM_A_BA_L, lower_32_bits(table->table.map));
2441 	hr_reg_write(r_a, CFG_LLM_A_BA_H, upper_32_bits(table->table.map));
2442 	hr_reg_write(r_a, CFG_LLM_A_DEPTH, buf->npages);
2443 	hr_reg_write(r_a, CFG_LLM_A_PGSZ, to_hr_hw_page_shift(buf->page_shift));
2444 	hr_reg_enable(r_a, CFG_LLM_A_INIT_EN);
2445 
2446 	addr = to_hr_hw_page_addr(hns_roce_buf_page(buf, 0));
2447 	hr_reg_write(r_a, CFG_LLM_A_HEAD_BA_L, lower_32_bits(addr));
2448 	hr_reg_write(r_a, CFG_LLM_A_HEAD_BA_H, upper_32_bits(addr));
2449 	hr_reg_write(r_a, CFG_LLM_A_HEAD_NXTPTR, 1);
2450 	hr_reg_write(r_a, CFG_LLM_A_HEAD_PTR, 0);
2451 
2452 	addr = to_hr_hw_page_addr(hns_roce_buf_page(buf, buf->npages - 1));
2453 	hr_reg_write(r_b, CFG_LLM_B_TAIL_BA_L, lower_32_bits(addr));
2454 	hr_reg_write(r_b, CFG_LLM_B_TAIL_BA_H, upper_32_bits(addr));
2455 	hr_reg_write(r_b, CFG_LLM_B_TAIL_PTR, buf->npages - 1);
2456 
2457 	return hns_roce_cmq_send(hr_dev, desc, 2);
2458 }
2459 
2460 static struct hns_roce_link_table *
2461 alloc_link_table_buf(struct hns_roce_dev *hr_dev)
2462 {
2463 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2464 	struct hns_roce_link_table *link_tbl;
2465 	u32 pg_shift, size, min_size;
2466 
2467 	link_tbl = &priv->ext_llm;
2468 	pg_shift = hr_dev->caps.llm_buf_pg_sz + PAGE_SHIFT;
2469 	size = hr_dev->caps.num_qps * HNS_ROCE_V2_EXT_LLM_ENTRY_SZ;
2470 	min_size = HNS_ROCE_EXT_LLM_MIN_PAGES(hr_dev->caps.sl_num) << pg_shift;
2471 
2472 	/* Alloc data table */
2473 	size = max(size, min_size);
2474 	link_tbl->buf = hns_roce_buf_alloc(hr_dev, size, pg_shift, 0);
2475 	if (IS_ERR(link_tbl->buf))
2476 		return ERR_PTR(-ENOMEM);
2477 
2478 	/* Alloc config table */
2479 	size = link_tbl->buf->npages * sizeof(u64);
2480 	link_tbl->table.buf = dma_alloc_coherent(hr_dev->dev, size,
2481 						 &link_tbl->table.map,
2482 						 GFP_KERNEL);
2483 	if (!link_tbl->table.buf) {
2484 		hns_roce_buf_free(hr_dev, link_tbl->buf);
2485 		return ERR_PTR(-ENOMEM);
2486 	}
2487 
2488 	return link_tbl;
2489 }
2490 
2491 static void free_link_table_buf(struct hns_roce_dev *hr_dev,
2492 				struct hns_roce_link_table *tbl)
2493 {
2494 	if (tbl->buf) {
2495 		u32 size = tbl->buf->npages * sizeof(u64);
2496 
2497 		dma_free_coherent(hr_dev->dev, size, tbl->table.buf,
2498 				  tbl->table.map);
2499 	}
2500 
2501 	hns_roce_buf_free(hr_dev, tbl->buf);
2502 }
2503 
2504 static int hns_roce_init_link_table(struct hns_roce_dev *hr_dev)
2505 {
2506 	struct hns_roce_link_table *link_tbl;
2507 	int ret;
2508 
2509 	link_tbl = alloc_link_table_buf(hr_dev);
2510 	if (IS_ERR(link_tbl))
2511 		return -ENOMEM;
2512 
2513 	if (WARN_ON(link_tbl->buf->npages > HNS_ROCE_V2_EXT_LLM_MAX_DEPTH)) {
2514 		ret = -EINVAL;
2515 		goto err_alloc;
2516 	}
2517 
2518 	config_llm_table(link_tbl->buf, link_tbl->table.buf);
2519 	ret = set_llm_cfg_to_hw(hr_dev, link_tbl);
2520 	if (ret)
2521 		goto err_alloc;
2522 
2523 	return 0;
2524 
2525 err_alloc:
2526 	free_link_table_buf(hr_dev, link_tbl);
2527 	return ret;
2528 }
2529 
2530 static void hns_roce_free_link_table(struct hns_roce_dev *hr_dev)
2531 {
2532 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2533 
2534 	free_link_table_buf(hr_dev, &priv->ext_llm);
2535 }
2536 
2537 static void free_dip_list(struct hns_roce_dev *hr_dev)
2538 {
2539 	struct hns_roce_dip *hr_dip;
2540 	struct hns_roce_dip *tmp;
2541 	unsigned long flags;
2542 
2543 	spin_lock_irqsave(&hr_dev->dip_list_lock, flags);
2544 
2545 	list_for_each_entry_safe(hr_dip, tmp, &hr_dev->dip_list, node) {
2546 		list_del(&hr_dip->node);
2547 		kfree(hr_dip);
2548 	}
2549 
2550 	spin_unlock_irqrestore(&hr_dev->dip_list_lock, flags);
2551 }
2552 
2553 static struct ib_pd *free_mr_init_pd(struct hns_roce_dev *hr_dev)
2554 {
2555 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2556 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2557 	struct ib_device *ibdev = &hr_dev->ib_dev;
2558 	struct hns_roce_pd *hr_pd;
2559 	struct ib_pd *pd;
2560 
2561 	hr_pd = kzalloc(sizeof(*hr_pd), GFP_KERNEL);
2562 	if (ZERO_OR_NULL_PTR(hr_pd))
2563 		return NULL;
2564 	pd = &hr_pd->ibpd;
2565 	pd->device = ibdev;
2566 
2567 	if (hns_roce_alloc_pd(pd, NULL)) {
2568 		ibdev_err(ibdev, "failed to create pd for free mr.\n");
2569 		kfree(hr_pd);
2570 		return NULL;
2571 	}
2572 	free_mr->rsv_pd = to_hr_pd(pd);
2573 	free_mr->rsv_pd->ibpd.device = &hr_dev->ib_dev;
2574 	free_mr->rsv_pd->ibpd.uobject = NULL;
2575 	free_mr->rsv_pd->ibpd.__internal_mr = NULL;
2576 	atomic_set(&free_mr->rsv_pd->ibpd.usecnt, 0);
2577 
2578 	return pd;
2579 }
2580 
2581 static struct ib_cq *free_mr_init_cq(struct hns_roce_dev *hr_dev)
2582 {
2583 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2584 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2585 	struct ib_device *ibdev = &hr_dev->ib_dev;
2586 	struct ib_cq_init_attr cq_init_attr = {};
2587 	struct hns_roce_cq *hr_cq;
2588 	struct ib_cq *cq;
2589 
2590 	cq_init_attr.cqe = HNS_ROCE_FREE_MR_USED_CQE_NUM;
2591 
2592 	hr_cq = kzalloc(sizeof(*hr_cq), GFP_KERNEL);
2593 	if (ZERO_OR_NULL_PTR(hr_cq))
2594 		return NULL;
2595 
2596 	cq = &hr_cq->ib_cq;
2597 	cq->device = ibdev;
2598 
2599 	if (hns_roce_create_cq(cq, &cq_init_attr, NULL)) {
2600 		ibdev_err(ibdev, "failed to create cq for free mr.\n");
2601 		kfree(hr_cq);
2602 		return NULL;
2603 	}
2604 	free_mr->rsv_cq = to_hr_cq(cq);
2605 	free_mr->rsv_cq->ib_cq.device = &hr_dev->ib_dev;
2606 	free_mr->rsv_cq->ib_cq.uobject = NULL;
2607 	free_mr->rsv_cq->ib_cq.comp_handler = NULL;
2608 	free_mr->rsv_cq->ib_cq.event_handler = NULL;
2609 	free_mr->rsv_cq->ib_cq.cq_context = NULL;
2610 	atomic_set(&free_mr->rsv_cq->ib_cq.usecnt, 0);
2611 
2612 	return cq;
2613 }
2614 
2615 static int free_mr_init_qp(struct hns_roce_dev *hr_dev, struct ib_cq *cq,
2616 			   struct ib_qp_init_attr *init_attr, int i)
2617 {
2618 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2619 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2620 	struct ib_device *ibdev = &hr_dev->ib_dev;
2621 	struct hns_roce_qp *hr_qp;
2622 	struct ib_qp *qp;
2623 	int ret;
2624 
2625 	hr_qp = kzalloc(sizeof(*hr_qp), GFP_KERNEL);
2626 	if (ZERO_OR_NULL_PTR(hr_qp))
2627 		return -ENOMEM;
2628 
2629 	qp = &hr_qp->ibqp;
2630 	qp->device = ibdev;
2631 
2632 	ret = hns_roce_create_qp(qp, init_attr, NULL);
2633 	if (ret) {
2634 		ibdev_err(ibdev, "failed to create qp for free mr.\n");
2635 		kfree(hr_qp);
2636 		return ret;
2637 	}
2638 
2639 	free_mr->rsv_qp[i] = hr_qp;
2640 	free_mr->rsv_qp[i]->ibqp.recv_cq = cq;
2641 	free_mr->rsv_qp[i]->ibqp.send_cq = cq;
2642 
2643 	return 0;
2644 }
2645 
2646 static void free_mr_exit(struct hns_roce_dev *hr_dev)
2647 {
2648 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2649 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2650 	struct ib_qp *qp;
2651 	int i;
2652 
2653 	for (i = 0; i < ARRAY_SIZE(free_mr->rsv_qp); i++) {
2654 		if (free_mr->rsv_qp[i]) {
2655 			qp = &free_mr->rsv_qp[i]->ibqp;
2656 			hns_roce_v2_destroy_qp(qp, NULL);
2657 			kfree(free_mr->rsv_qp[i]);
2658 			free_mr->rsv_qp[i] = NULL;
2659 		}
2660 	}
2661 
2662 	if (free_mr->rsv_cq) {
2663 		hns_roce_destroy_cq(&free_mr->rsv_cq->ib_cq, NULL);
2664 		kfree(free_mr->rsv_cq);
2665 		free_mr->rsv_cq = NULL;
2666 	}
2667 
2668 	if (free_mr->rsv_pd) {
2669 		hns_roce_dealloc_pd(&free_mr->rsv_pd->ibpd, NULL);
2670 		kfree(free_mr->rsv_pd);
2671 		free_mr->rsv_pd = NULL;
2672 	}
2673 }
2674 
2675 static int free_mr_alloc_res(struct hns_roce_dev *hr_dev)
2676 {
2677 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2678 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2679 	struct ib_qp_init_attr qp_init_attr = {};
2680 	struct ib_pd *pd;
2681 	struct ib_cq *cq;
2682 	int ret;
2683 	int i;
2684 
2685 	pd = free_mr_init_pd(hr_dev);
2686 	if (!pd)
2687 		return -ENOMEM;
2688 
2689 	cq = free_mr_init_cq(hr_dev);
2690 	if (!cq) {
2691 		ret = -ENOMEM;
2692 		goto create_failed_cq;
2693 	}
2694 
2695 	qp_init_attr.qp_type = IB_QPT_RC;
2696 	qp_init_attr.sq_sig_type = IB_SIGNAL_ALL_WR;
2697 	qp_init_attr.send_cq = cq;
2698 	qp_init_attr.recv_cq = cq;
2699 	for (i = 0; i < ARRAY_SIZE(free_mr->rsv_qp); i++) {
2700 		qp_init_attr.cap.max_send_wr = HNS_ROCE_FREE_MR_USED_SQWQE_NUM;
2701 		qp_init_attr.cap.max_send_sge = HNS_ROCE_FREE_MR_USED_SQSGE_NUM;
2702 		qp_init_attr.cap.max_recv_wr = HNS_ROCE_FREE_MR_USED_RQWQE_NUM;
2703 		qp_init_attr.cap.max_recv_sge = HNS_ROCE_FREE_MR_USED_RQSGE_NUM;
2704 
2705 		ret = free_mr_init_qp(hr_dev, cq, &qp_init_attr, i);
2706 		if (ret)
2707 			goto create_failed_qp;
2708 	}
2709 
2710 	return 0;
2711 
2712 create_failed_qp:
2713 	for (i--; i >= 0; i--) {
2714 		hns_roce_v2_destroy_qp(&free_mr->rsv_qp[i]->ibqp, NULL);
2715 		kfree(free_mr->rsv_qp[i]);
2716 	}
2717 	hns_roce_destroy_cq(cq, NULL);
2718 	kfree(cq);
2719 
2720 create_failed_cq:
2721 	hns_roce_dealloc_pd(pd, NULL);
2722 	kfree(pd);
2723 
2724 	return ret;
2725 }
2726 
2727 static int free_mr_modify_rsv_qp(struct hns_roce_dev *hr_dev,
2728 				 struct ib_qp_attr *attr, int sl_num)
2729 {
2730 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2731 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2732 	struct ib_device *ibdev = &hr_dev->ib_dev;
2733 	struct hns_roce_qp *hr_qp;
2734 	int loopback;
2735 	int mask;
2736 	int ret;
2737 
2738 	hr_qp = to_hr_qp(&free_mr->rsv_qp[sl_num]->ibqp);
2739 	hr_qp->free_mr_en = 1;
2740 	hr_qp->ibqp.device = ibdev;
2741 	hr_qp->ibqp.qp_type = IB_QPT_RC;
2742 
2743 	mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT | IB_QP_ACCESS_FLAGS;
2744 	attr->qp_state = IB_QPS_INIT;
2745 	attr->port_num = 1;
2746 	attr->qp_access_flags = IB_ACCESS_REMOTE_WRITE;
2747 	ret = hr_dev->hw->modify_qp(&hr_qp->ibqp, attr, mask, IB_QPS_INIT,
2748 				    IB_QPS_INIT, NULL);
2749 	if (ret) {
2750 		ibdev_err(ibdev, "failed to modify qp to init, ret = %d.\n",
2751 			  ret);
2752 		return ret;
2753 	}
2754 
2755 	loopback = hr_dev->loop_idc;
2756 	/* Set qpc lbi = 1 incidate loopback IO */
2757 	hr_dev->loop_idc = 1;
2758 
2759 	mask = IB_QP_STATE | IB_QP_AV | IB_QP_PATH_MTU | IB_QP_DEST_QPN |
2760 	       IB_QP_RQ_PSN | IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_MIN_RNR_TIMER;
2761 	attr->qp_state = IB_QPS_RTR;
2762 	attr->ah_attr.type = RDMA_AH_ATTR_TYPE_ROCE;
2763 	attr->path_mtu = IB_MTU_256;
2764 	attr->dest_qp_num = hr_qp->qpn;
2765 	attr->rq_psn = HNS_ROCE_FREE_MR_USED_PSN;
2766 
2767 	rdma_ah_set_sl(&attr->ah_attr, (u8)sl_num);
2768 
2769 	ret = hr_dev->hw->modify_qp(&hr_qp->ibqp, attr, mask, IB_QPS_INIT,
2770 				    IB_QPS_RTR, NULL);
2771 	hr_dev->loop_idc = loopback;
2772 	if (ret) {
2773 		ibdev_err(ibdev, "failed to modify qp to rtr, ret = %d.\n",
2774 			  ret);
2775 		return ret;
2776 	}
2777 
2778 	mask = IB_QP_STATE | IB_QP_SQ_PSN | IB_QP_RETRY_CNT | IB_QP_TIMEOUT |
2779 	       IB_QP_RNR_RETRY | IB_QP_MAX_QP_RD_ATOMIC;
2780 	attr->qp_state = IB_QPS_RTS;
2781 	attr->sq_psn = HNS_ROCE_FREE_MR_USED_PSN;
2782 	attr->retry_cnt = HNS_ROCE_FREE_MR_USED_QP_RETRY_CNT;
2783 	attr->timeout = HNS_ROCE_FREE_MR_USED_QP_TIMEOUT;
2784 	ret = hr_dev->hw->modify_qp(&hr_qp->ibqp, attr, mask, IB_QPS_RTR,
2785 				    IB_QPS_RTS, NULL);
2786 	if (ret)
2787 		ibdev_err(ibdev, "failed to modify qp to rts, ret = %d.\n",
2788 			  ret);
2789 
2790 	return ret;
2791 }
2792 
2793 static int free_mr_modify_qp(struct hns_roce_dev *hr_dev)
2794 {
2795 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2796 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2797 	struct ib_qp_attr attr = {};
2798 	int ret;
2799 	int i;
2800 
2801 	rdma_ah_set_grh(&attr.ah_attr, NULL, 0, 0, 1, 0);
2802 	rdma_ah_set_static_rate(&attr.ah_attr, 3);
2803 	rdma_ah_set_port_num(&attr.ah_attr, 1);
2804 
2805 	for (i = 0; i < ARRAY_SIZE(free_mr->rsv_qp); i++) {
2806 		ret = free_mr_modify_rsv_qp(hr_dev, &attr, i);
2807 		if (ret)
2808 			return ret;
2809 	}
2810 
2811 	return 0;
2812 }
2813 
2814 static int free_mr_init(struct hns_roce_dev *hr_dev)
2815 {
2816 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2817 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2818 	int ret;
2819 
2820 	mutex_init(&free_mr->mutex);
2821 
2822 	ret = free_mr_alloc_res(hr_dev);
2823 	if (ret)
2824 		return ret;
2825 
2826 	ret = free_mr_modify_qp(hr_dev);
2827 	if (ret)
2828 		goto err_modify_qp;
2829 
2830 	return 0;
2831 
2832 err_modify_qp:
2833 	free_mr_exit(hr_dev);
2834 
2835 	return ret;
2836 }
2837 
2838 static int get_hem_table(struct hns_roce_dev *hr_dev)
2839 {
2840 	unsigned int qpc_count;
2841 	unsigned int cqc_count;
2842 	unsigned int gmv_count;
2843 	int ret;
2844 	int i;
2845 
2846 	/* Alloc memory for source address table buffer space chunk */
2847 	for (gmv_count = 0; gmv_count < hr_dev->caps.gmv_entry_num;
2848 	     gmv_count++) {
2849 		ret = hns_roce_table_get(hr_dev, &hr_dev->gmv_table, gmv_count);
2850 		if (ret)
2851 			goto err_gmv_failed;
2852 	}
2853 
2854 	if (hr_dev->is_vf)
2855 		return 0;
2856 
2857 	/* Alloc memory for QPC Timer buffer space chunk */
2858 	for (qpc_count = 0; qpc_count < hr_dev->caps.qpc_timer_bt_num;
2859 	     qpc_count++) {
2860 		ret = hns_roce_table_get(hr_dev, &hr_dev->qpc_timer_table,
2861 					 qpc_count);
2862 		if (ret) {
2863 			dev_err(hr_dev->dev, "QPC Timer get failed\n");
2864 			goto err_qpc_timer_failed;
2865 		}
2866 	}
2867 
2868 	/* Alloc memory for CQC Timer buffer space chunk */
2869 	for (cqc_count = 0; cqc_count < hr_dev->caps.cqc_timer_bt_num;
2870 	     cqc_count++) {
2871 		ret = hns_roce_table_get(hr_dev, &hr_dev->cqc_timer_table,
2872 					 cqc_count);
2873 		if (ret) {
2874 			dev_err(hr_dev->dev, "CQC Timer get failed\n");
2875 			goto err_cqc_timer_failed;
2876 		}
2877 	}
2878 
2879 	return 0;
2880 
2881 err_cqc_timer_failed:
2882 	for (i = 0; i < cqc_count; i++)
2883 		hns_roce_table_put(hr_dev, &hr_dev->cqc_timer_table, i);
2884 
2885 err_qpc_timer_failed:
2886 	for (i = 0; i < qpc_count; i++)
2887 		hns_roce_table_put(hr_dev, &hr_dev->qpc_timer_table, i);
2888 
2889 err_gmv_failed:
2890 	for (i = 0; i < gmv_count; i++)
2891 		hns_roce_table_put(hr_dev, &hr_dev->gmv_table, i);
2892 
2893 	return ret;
2894 }
2895 
2896 static void put_hem_table(struct hns_roce_dev *hr_dev)
2897 {
2898 	int i;
2899 
2900 	for (i = 0; i < hr_dev->caps.gmv_entry_num; i++)
2901 		hns_roce_table_put(hr_dev, &hr_dev->gmv_table, i);
2902 
2903 	if (hr_dev->is_vf)
2904 		return;
2905 
2906 	for (i = 0; i < hr_dev->caps.qpc_timer_bt_num; i++)
2907 		hns_roce_table_put(hr_dev, &hr_dev->qpc_timer_table, i);
2908 
2909 	for (i = 0; i < hr_dev->caps.cqc_timer_bt_num; i++)
2910 		hns_roce_table_put(hr_dev, &hr_dev->cqc_timer_table, i);
2911 }
2912 
2913 static int hns_roce_v2_init(struct hns_roce_dev *hr_dev)
2914 {
2915 	int ret;
2916 
2917 	/* The hns ROCEE requires the extdb info to be cleared before using */
2918 	ret = hns_roce_clear_extdb_list_info(hr_dev);
2919 	if (ret)
2920 		return ret;
2921 
2922 	ret = get_hem_table(hr_dev);
2923 	if (ret)
2924 		return ret;
2925 
2926 	if (hr_dev->is_vf)
2927 		return 0;
2928 
2929 	ret = hns_roce_init_link_table(hr_dev);
2930 	if (ret) {
2931 		dev_err(hr_dev->dev, "failed to init llm, ret = %d.\n", ret);
2932 		goto err_llm_init_failed;
2933 	}
2934 
2935 	return 0;
2936 
2937 err_llm_init_failed:
2938 	put_hem_table(hr_dev);
2939 
2940 	return ret;
2941 }
2942 
2943 static void hns_roce_v2_exit(struct hns_roce_dev *hr_dev)
2944 {
2945 	hns_roce_function_clear(hr_dev);
2946 
2947 	if (!hr_dev->is_vf)
2948 		hns_roce_free_link_table(hr_dev);
2949 
2950 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP09)
2951 		free_dip_list(hr_dev);
2952 }
2953 
2954 static int hns_roce_mbox_post(struct hns_roce_dev *hr_dev,
2955 			      struct hns_roce_mbox_msg *mbox_msg)
2956 {
2957 	struct hns_roce_cmq_desc desc;
2958 	struct hns_roce_post_mbox *mb = (struct hns_roce_post_mbox *)desc.data;
2959 
2960 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_POST_MB, false);
2961 
2962 	mb->in_param_l = cpu_to_le32(mbox_msg->in_param);
2963 	mb->in_param_h = cpu_to_le32(mbox_msg->in_param >> 32);
2964 	mb->out_param_l = cpu_to_le32(mbox_msg->out_param);
2965 	mb->out_param_h = cpu_to_le32(mbox_msg->out_param >> 32);
2966 	mb->cmd_tag = cpu_to_le32(mbox_msg->tag << 8 | mbox_msg->cmd);
2967 	mb->token_event_en = cpu_to_le32(mbox_msg->event_en << 16 |
2968 					 mbox_msg->token);
2969 
2970 	return hns_roce_cmq_send(hr_dev, &desc, 1);
2971 }
2972 
2973 static int v2_wait_mbox_complete(struct hns_roce_dev *hr_dev, u32 timeout,
2974 				 u8 *complete_status)
2975 {
2976 	struct hns_roce_mbox_status *mb_st;
2977 	struct hns_roce_cmq_desc desc;
2978 	unsigned long end;
2979 	int ret = -EBUSY;
2980 	u32 status;
2981 	bool busy;
2982 
2983 	mb_st = (struct hns_roce_mbox_status *)desc.data;
2984 	end = msecs_to_jiffies(timeout) + jiffies;
2985 	while (v2_chk_mbox_is_avail(hr_dev, &busy)) {
2986 		if (hr_dev->cmd.state == HNS_ROCE_CMDQ_STATE_FATAL_ERR)
2987 			return -EIO;
2988 
2989 		status = 0;
2990 		hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_MB_ST,
2991 					      true);
2992 		ret = __hns_roce_cmq_send(hr_dev, &desc, 1);
2993 		if (!ret) {
2994 			status = le32_to_cpu(mb_st->mb_status_hw_run);
2995 			/* No pending message exists in ROCEE mbox. */
2996 			if (!(status & MB_ST_HW_RUN_M))
2997 				break;
2998 		} else if (!v2_chk_mbox_is_avail(hr_dev, &busy)) {
2999 			break;
3000 		}
3001 
3002 		if (time_after(jiffies, end)) {
3003 			dev_err_ratelimited(hr_dev->dev,
3004 					    "failed to wait mbox status 0x%x\n",
3005 					    status);
3006 			return -ETIMEDOUT;
3007 		}
3008 
3009 		cond_resched();
3010 		ret = -EBUSY;
3011 	}
3012 
3013 	if (!ret) {
3014 		*complete_status = (u8)(status & MB_ST_COMPLETE_M);
3015 	} else if (!v2_chk_mbox_is_avail(hr_dev, &busy)) {
3016 		/* Ignore all errors if the mbox is unavailable. */
3017 		ret = 0;
3018 		*complete_status = MB_ST_COMPLETE_M;
3019 	}
3020 
3021 	return ret;
3022 }
3023 
3024 static int v2_post_mbox(struct hns_roce_dev *hr_dev,
3025 			struct hns_roce_mbox_msg *mbox_msg)
3026 {
3027 	u8 status = 0;
3028 	int ret;
3029 
3030 	/* Waiting for the mbox to be idle */
3031 	ret = v2_wait_mbox_complete(hr_dev, HNS_ROCE_V2_GO_BIT_TIMEOUT_MSECS,
3032 				    &status);
3033 	if (unlikely(ret)) {
3034 		dev_err_ratelimited(hr_dev->dev,
3035 				    "failed to check post mbox status = 0x%x, ret = %d.\n",
3036 				    status, ret);
3037 		return ret;
3038 	}
3039 
3040 	/* Post new message to mbox */
3041 	ret = hns_roce_mbox_post(hr_dev, mbox_msg);
3042 	if (ret)
3043 		dev_err_ratelimited(hr_dev->dev,
3044 				    "failed to post mailbox, ret = %d.\n", ret);
3045 
3046 	return ret;
3047 }
3048 
3049 static int v2_poll_mbox_done(struct hns_roce_dev *hr_dev)
3050 {
3051 	u8 status = 0;
3052 	int ret;
3053 
3054 	ret = v2_wait_mbox_complete(hr_dev, HNS_ROCE_CMD_TIMEOUT_MSECS,
3055 				    &status);
3056 	if (!ret) {
3057 		if (status != MB_ST_COMPLETE_SUCC)
3058 			return -EBUSY;
3059 	} else {
3060 		dev_err_ratelimited(hr_dev->dev,
3061 				    "failed to check mbox status = 0x%x, ret = %d.\n",
3062 				    status, ret);
3063 	}
3064 
3065 	return ret;
3066 }
3067 
3068 static void copy_gid(void *dest, const union ib_gid *gid)
3069 {
3070 #define GID_SIZE 4
3071 	const union ib_gid *src = gid;
3072 	__le32 (*p)[GID_SIZE] = dest;
3073 	int i;
3074 
3075 	if (!gid)
3076 		src = &zgid;
3077 
3078 	for (i = 0; i < GID_SIZE; i++)
3079 		(*p)[i] = cpu_to_le32(*(u32 *)&src->raw[i * sizeof(u32)]);
3080 }
3081 
3082 static int config_sgid_table(struct hns_roce_dev *hr_dev,
3083 			     int gid_index, const union ib_gid *gid,
3084 			     enum hns_roce_sgid_type sgid_type)
3085 {
3086 	struct hns_roce_cmq_desc desc;
3087 	struct hns_roce_cfg_sgid_tb *sgid_tb =
3088 				    (struct hns_roce_cfg_sgid_tb *)desc.data;
3089 
3090 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_SGID_TB, false);
3091 
3092 	hr_reg_write(sgid_tb, CFG_SGID_TB_TABLE_IDX, gid_index);
3093 	hr_reg_write(sgid_tb, CFG_SGID_TB_VF_SGID_TYPE, sgid_type);
3094 
3095 	copy_gid(&sgid_tb->vf_sgid_l, gid);
3096 
3097 	return hns_roce_cmq_send(hr_dev, &desc, 1);
3098 }
3099 
3100 static int config_gmv_table(struct hns_roce_dev *hr_dev,
3101 			    int gid_index, const union ib_gid *gid,
3102 			    enum hns_roce_sgid_type sgid_type,
3103 			    const struct ib_gid_attr *attr)
3104 {
3105 	struct hns_roce_cmq_desc desc[2];
3106 	struct hns_roce_cfg_gmv_tb_a *tb_a =
3107 				(struct hns_roce_cfg_gmv_tb_a *)desc[0].data;
3108 	struct hns_roce_cfg_gmv_tb_b *tb_b =
3109 				(struct hns_roce_cfg_gmv_tb_b *)desc[1].data;
3110 
3111 	u16 vlan_id = VLAN_CFI_MASK;
3112 	u8 mac[ETH_ALEN] = {};
3113 	int ret;
3114 
3115 	if (gid) {
3116 		ret = rdma_read_gid_l2_fields(attr, &vlan_id, mac);
3117 		if (ret)
3118 			return ret;
3119 	}
3120 
3121 	hns_roce_cmq_setup_basic_desc(&desc[0], HNS_ROCE_OPC_CFG_GMV_TBL, false);
3122 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
3123 
3124 	hns_roce_cmq_setup_basic_desc(&desc[1], HNS_ROCE_OPC_CFG_GMV_TBL, false);
3125 
3126 	copy_gid(&tb_a->vf_sgid_l, gid);
3127 
3128 	hr_reg_write(tb_a, GMV_TB_A_VF_SGID_TYPE, sgid_type);
3129 	hr_reg_write(tb_a, GMV_TB_A_VF_VLAN_EN, vlan_id < VLAN_CFI_MASK);
3130 	hr_reg_write(tb_a, GMV_TB_A_VF_VLAN_ID, vlan_id);
3131 
3132 	tb_b->vf_smac_l = cpu_to_le32(*(u32 *)mac);
3133 
3134 	hr_reg_write(tb_b, GMV_TB_B_SMAC_H, *(u16 *)&mac[4]);
3135 	hr_reg_write(tb_b, GMV_TB_B_SGID_IDX, gid_index);
3136 
3137 	return hns_roce_cmq_send(hr_dev, desc, 2);
3138 }
3139 
3140 static int hns_roce_v2_set_gid(struct hns_roce_dev *hr_dev, int gid_index,
3141 			       const union ib_gid *gid,
3142 			       const struct ib_gid_attr *attr)
3143 {
3144 	enum hns_roce_sgid_type sgid_type = GID_TYPE_FLAG_ROCE_V1;
3145 	int ret;
3146 
3147 	if (gid) {
3148 		if (attr->gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
3149 			if (ipv6_addr_v4mapped((void *)gid))
3150 				sgid_type = GID_TYPE_FLAG_ROCE_V2_IPV4;
3151 			else
3152 				sgid_type = GID_TYPE_FLAG_ROCE_V2_IPV6;
3153 		} else if (attr->gid_type == IB_GID_TYPE_ROCE) {
3154 			sgid_type = GID_TYPE_FLAG_ROCE_V1;
3155 		}
3156 	}
3157 
3158 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
3159 		ret = config_gmv_table(hr_dev, gid_index, gid, sgid_type, attr);
3160 	else
3161 		ret = config_sgid_table(hr_dev, gid_index, gid, sgid_type);
3162 
3163 	if (ret)
3164 		ibdev_err(&hr_dev->ib_dev, "failed to set gid, ret = %d!\n",
3165 			  ret);
3166 
3167 	return ret;
3168 }
3169 
3170 static int hns_roce_v2_set_mac(struct hns_roce_dev *hr_dev, u8 phy_port,
3171 			       const u8 *addr)
3172 {
3173 	struct hns_roce_cmq_desc desc;
3174 	struct hns_roce_cfg_smac_tb *smac_tb =
3175 				    (struct hns_roce_cfg_smac_tb *)desc.data;
3176 	u16 reg_smac_h;
3177 	u32 reg_smac_l;
3178 
3179 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_SMAC_TB, false);
3180 
3181 	reg_smac_l = *(u32 *)(&addr[0]);
3182 	reg_smac_h = *(u16 *)(&addr[4]);
3183 
3184 	hr_reg_write(smac_tb, CFG_SMAC_TB_IDX, phy_port);
3185 	hr_reg_write(smac_tb, CFG_SMAC_TB_VF_SMAC_H, reg_smac_h);
3186 	smac_tb->vf_smac_l = cpu_to_le32(reg_smac_l);
3187 
3188 	return hns_roce_cmq_send(hr_dev, &desc, 1);
3189 }
3190 
3191 static int set_mtpt_pbl(struct hns_roce_dev *hr_dev,
3192 			struct hns_roce_v2_mpt_entry *mpt_entry,
3193 			struct hns_roce_mr *mr)
3194 {
3195 	u64 pages[HNS_ROCE_V2_MAX_INNER_MTPT_NUM] = { 0 };
3196 	struct ib_device *ibdev = &hr_dev->ib_dev;
3197 	dma_addr_t pbl_ba;
3198 	int i, count;
3199 
3200 	count = hns_roce_mtr_find(hr_dev, &mr->pbl_mtr, 0, pages,
3201 				  min_t(int, ARRAY_SIZE(pages), mr->npages),
3202 				  &pbl_ba);
3203 	if (count < 1) {
3204 		ibdev_err(ibdev, "failed to find PBL mtr, count = %d.\n",
3205 			  count);
3206 		return -ENOBUFS;
3207 	}
3208 
3209 	/* Aligned to the hardware address access unit */
3210 	for (i = 0; i < count; i++)
3211 		pages[i] >>= 6;
3212 
3213 	mpt_entry->pbl_size = cpu_to_le32(mr->npages);
3214 	mpt_entry->pbl_ba_l = cpu_to_le32(pbl_ba >> 3);
3215 	hr_reg_write(mpt_entry, MPT_PBL_BA_H, upper_32_bits(pbl_ba >> 3));
3216 
3217 	mpt_entry->pa0_l = cpu_to_le32(lower_32_bits(pages[0]));
3218 	hr_reg_write(mpt_entry, MPT_PA0_H, upper_32_bits(pages[0]));
3219 
3220 	mpt_entry->pa1_l = cpu_to_le32(lower_32_bits(pages[1]));
3221 	hr_reg_write(mpt_entry, MPT_PA1_H, upper_32_bits(pages[1]));
3222 	hr_reg_write(mpt_entry, MPT_PBL_BUF_PG_SZ,
3223 		     to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.buf_pg_shift));
3224 
3225 	return 0;
3226 }
3227 
3228 static int hns_roce_v2_write_mtpt(struct hns_roce_dev *hr_dev,
3229 				  void *mb_buf, struct hns_roce_mr *mr)
3230 {
3231 	struct hns_roce_v2_mpt_entry *mpt_entry;
3232 
3233 	mpt_entry = mb_buf;
3234 	memset(mpt_entry, 0, sizeof(*mpt_entry));
3235 
3236 	hr_reg_write(mpt_entry, MPT_ST, V2_MPT_ST_VALID);
3237 	hr_reg_write(mpt_entry, MPT_PD, mr->pd);
3238 
3239 	hr_reg_write_bool(mpt_entry, MPT_BIND_EN,
3240 			  mr->access & IB_ACCESS_MW_BIND);
3241 	hr_reg_write_bool(mpt_entry, MPT_ATOMIC_EN,
3242 			  mr->access & IB_ACCESS_REMOTE_ATOMIC);
3243 	hr_reg_write_bool(mpt_entry, MPT_RR_EN,
3244 			  mr->access & IB_ACCESS_REMOTE_READ);
3245 	hr_reg_write_bool(mpt_entry, MPT_RW_EN,
3246 			  mr->access & IB_ACCESS_REMOTE_WRITE);
3247 	hr_reg_write_bool(mpt_entry, MPT_LW_EN,
3248 			  mr->access & IB_ACCESS_LOCAL_WRITE);
3249 
3250 	mpt_entry->len_l = cpu_to_le32(lower_32_bits(mr->size));
3251 	mpt_entry->len_h = cpu_to_le32(upper_32_bits(mr->size));
3252 	mpt_entry->lkey = cpu_to_le32(mr->key);
3253 	mpt_entry->va_l = cpu_to_le32(lower_32_bits(mr->iova));
3254 	mpt_entry->va_h = cpu_to_le32(upper_32_bits(mr->iova));
3255 
3256 	if (mr->type != MR_TYPE_MR)
3257 		hr_reg_enable(mpt_entry, MPT_PA);
3258 
3259 	if (mr->type == MR_TYPE_DMA)
3260 		return 0;
3261 
3262 	if (mr->pbl_hop_num != HNS_ROCE_HOP_NUM_0)
3263 		hr_reg_write(mpt_entry, MPT_PBL_HOP_NUM, mr->pbl_hop_num);
3264 
3265 	hr_reg_write(mpt_entry, MPT_PBL_BA_PG_SZ,
3266 		     to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.ba_pg_shift));
3267 	hr_reg_enable(mpt_entry, MPT_INNER_PA_VLD);
3268 
3269 	return set_mtpt_pbl(hr_dev, mpt_entry, mr);
3270 }
3271 
3272 static int hns_roce_v2_rereg_write_mtpt(struct hns_roce_dev *hr_dev,
3273 					struct hns_roce_mr *mr, int flags,
3274 					void *mb_buf)
3275 {
3276 	struct hns_roce_v2_mpt_entry *mpt_entry = mb_buf;
3277 	u32 mr_access_flags = mr->access;
3278 	int ret = 0;
3279 
3280 	hr_reg_write(mpt_entry, MPT_ST, V2_MPT_ST_VALID);
3281 	hr_reg_write(mpt_entry, MPT_PD, mr->pd);
3282 
3283 	if (flags & IB_MR_REREG_ACCESS) {
3284 		hr_reg_write(mpt_entry, MPT_BIND_EN,
3285 			     (mr_access_flags & IB_ACCESS_MW_BIND ? 1 : 0));
3286 		hr_reg_write(mpt_entry, MPT_ATOMIC_EN,
3287 			     mr_access_flags & IB_ACCESS_REMOTE_ATOMIC ? 1 : 0);
3288 		hr_reg_write(mpt_entry, MPT_RR_EN,
3289 			     mr_access_flags & IB_ACCESS_REMOTE_READ ? 1 : 0);
3290 		hr_reg_write(mpt_entry, MPT_RW_EN,
3291 			     mr_access_flags & IB_ACCESS_REMOTE_WRITE ? 1 : 0);
3292 		hr_reg_write(mpt_entry, MPT_LW_EN,
3293 			     mr_access_flags & IB_ACCESS_LOCAL_WRITE ? 1 : 0);
3294 	}
3295 
3296 	if (flags & IB_MR_REREG_TRANS) {
3297 		mpt_entry->va_l = cpu_to_le32(lower_32_bits(mr->iova));
3298 		mpt_entry->va_h = cpu_to_le32(upper_32_bits(mr->iova));
3299 		mpt_entry->len_l = cpu_to_le32(lower_32_bits(mr->size));
3300 		mpt_entry->len_h = cpu_to_le32(upper_32_bits(mr->size));
3301 
3302 		ret = set_mtpt_pbl(hr_dev, mpt_entry, mr);
3303 	}
3304 
3305 	return ret;
3306 }
3307 
3308 static int hns_roce_v2_frmr_write_mtpt(struct hns_roce_dev *hr_dev,
3309 				       void *mb_buf, struct hns_roce_mr *mr)
3310 {
3311 	struct ib_device *ibdev = &hr_dev->ib_dev;
3312 	struct hns_roce_v2_mpt_entry *mpt_entry;
3313 	dma_addr_t pbl_ba = 0;
3314 
3315 	mpt_entry = mb_buf;
3316 	memset(mpt_entry, 0, sizeof(*mpt_entry));
3317 
3318 	if (hns_roce_mtr_find(hr_dev, &mr->pbl_mtr, 0, NULL, 0, &pbl_ba) < 0) {
3319 		ibdev_err(ibdev, "failed to find frmr mtr.\n");
3320 		return -ENOBUFS;
3321 	}
3322 
3323 	hr_reg_write(mpt_entry, MPT_ST, V2_MPT_ST_FREE);
3324 	hr_reg_write(mpt_entry, MPT_PD, mr->pd);
3325 
3326 	hr_reg_enable(mpt_entry, MPT_RA_EN);
3327 	hr_reg_enable(mpt_entry, MPT_R_INV_EN);
3328 
3329 	hr_reg_enable(mpt_entry, MPT_FRE);
3330 	hr_reg_clear(mpt_entry, MPT_MR_MW);
3331 	hr_reg_enable(mpt_entry, MPT_BPD);
3332 	hr_reg_clear(mpt_entry, MPT_PA);
3333 
3334 	hr_reg_write(mpt_entry, MPT_PBL_HOP_NUM, 1);
3335 	hr_reg_write(mpt_entry, MPT_PBL_BA_PG_SZ,
3336 		     to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.ba_pg_shift));
3337 	hr_reg_write(mpt_entry, MPT_PBL_BUF_PG_SZ,
3338 		     to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.buf_pg_shift));
3339 
3340 	mpt_entry->pbl_size = cpu_to_le32(mr->npages);
3341 
3342 	mpt_entry->pbl_ba_l = cpu_to_le32(lower_32_bits(pbl_ba >> 3));
3343 	hr_reg_write(mpt_entry, MPT_PBL_BA_H, upper_32_bits(pbl_ba >> 3));
3344 
3345 	return 0;
3346 }
3347 
3348 static int hns_roce_v2_mw_write_mtpt(void *mb_buf, struct hns_roce_mw *mw)
3349 {
3350 	struct hns_roce_v2_mpt_entry *mpt_entry;
3351 
3352 	mpt_entry = mb_buf;
3353 	memset(mpt_entry, 0, sizeof(*mpt_entry));
3354 
3355 	hr_reg_write(mpt_entry, MPT_ST, V2_MPT_ST_FREE);
3356 	hr_reg_write(mpt_entry, MPT_PD, mw->pdn);
3357 
3358 	hr_reg_enable(mpt_entry, MPT_R_INV_EN);
3359 	hr_reg_enable(mpt_entry, MPT_LW_EN);
3360 
3361 	hr_reg_enable(mpt_entry, MPT_MR_MW);
3362 	hr_reg_enable(mpt_entry, MPT_BPD);
3363 	hr_reg_clear(mpt_entry, MPT_PA);
3364 	hr_reg_write(mpt_entry, MPT_BQP,
3365 		     mw->ibmw.type == IB_MW_TYPE_1 ? 0 : 1);
3366 
3367 	mpt_entry->lkey = cpu_to_le32(mw->rkey);
3368 
3369 	hr_reg_write(mpt_entry, MPT_PBL_HOP_NUM,
3370 		     mw->pbl_hop_num == HNS_ROCE_HOP_NUM_0 ? 0 :
3371 							     mw->pbl_hop_num);
3372 	hr_reg_write(mpt_entry, MPT_PBL_BA_PG_SZ,
3373 		     mw->pbl_ba_pg_sz + PG_SHIFT_OFFSET);
3374 	hr_reg_write(mpt_entry, MPT_PBL_BUF_PG_SZ,
3375 		     mw->pbl_buf_pg_sz + PG_SHIFT_OFFSET);
3376 
3377 	return 0;
3378 }
3379 
3380 static int free_mr_post_send_lp_wqe(struct hns_roce_qp *hr_qp)
3381 {
3382 	struct hns_roce_dev *hr_dev = to_hr_dev(hr_qp->ibqp.device);
3383 	struct ib_device *ibdev = &hr_dev->ib_dev;
3384 	const struct ib_send_wr *bad_wr;
3385 	struct ib_rdma_wr rdma_wr = {};
3386 	struct ib_send_wr *send_wr;
3387 	int ret;
3388 
3389 	send_wr = &rdma_wr.wr;
3390 	send_wr->opcode = IB_WR_RDMA_WRITE;
3391 
3392 	ret = hns_roce_v2_post_send(&hr_qp->ibqp, send_wr, &bad_wr);
3393 	if (ret) {
3394 		ibdev_err(ibdev, "failed to post wqe for free mr, ret = %d.\n",
3395 			  ret);
3396 		return ret;
3397 	}
3398 
3399 	return 0;
3400 }
3401 
3402 static int hns_roce_v2_poll_cq(struct ib_cq *ibcq, int num_entries,
3403 			       struct ib_wc *wc);
3404 
3405 static void free_mr_send_cmd_to_hw(struct hns_roce_dev *hr_dev)
3406 {
3407 	struct hns_roce_v2_priv *priv = hr_dev->priv;
3408 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
3409 	struct ib_wc wc[ARRAY_SIZE(free_mr->rsv_qp)];
3410 	struct ib_device *ibdev = &hr_dev->ib_dev;
3411 	struct hns_roce_qp *hr_qp;
3412 	unsigned long end;
3413 	int cqe_cnt = 0;
3414 	int npolled;
3415 	int ret;
3416 	int i;
3417 
3418 	/*
3419 	 * If the device initialization is not complete or in the uninstall
3420 	 * process, then there is no need to execute free mr.
3421 	 */
3422 	if (priv->handle->rinfo.reset_state == HNS_ROCE_STATE_RST_INIT ||
3423 	    priv->handle->rinfo.instance_state == HNS_ROCE_STATE_INIT ||
3424 	    hr_dev->state == HNS_ROCE_DEVICE_STATE_UNINIT)
3425 		return;
3426 
3427 	mutex_lock(&free_mr->mutex);
3428 
3429 	for (i = 0; i < ARRAY_SIZE(free_mr->rsv_qp); i++) {
3430 		hr_qp = free_mr->rsv_qp[i];
3431 
3432 		ret = free_mr_post_send_lp_wqe(hr_qp);
3433 		if (ret) {
3434 			ibdev_err(ibdev,
3435 				  "failed to send wqe (qp:0x%lx) for free mr, ret = %d.\n",
3436 				  hr_qp->qpn, ret);
3437 			break;
3438 		}
3439 
3440 		cqe_cnt++;
3441 	}
3442 
3443 	end = msecs_to_jiffies(HNS_ROCE_V2_FREE_MR_TIMEOUT) + jiffies;
3444 	while (cqe_cnt) {
3445 		npolled = hns_roce_v2_poll_cq(&free_mr->rsv_cq->ib_cq, cqe_cnt, wc);
3446 		if (npolled < 0) {
3447 			ibdev_err(ibdev,
3448 				  "failed to poll cqe for free mr, remain %d cqe.\n",
3449 				  cqe_cnt);
3450 			goto out;
3451 		}
3452 
3453 		if (time_after(jiffies, end)) {
3454 			ibdev_err(ibdev,
3455 				  "failed to poll cqe for free mr and timeout, remain %d cqe.\n",
3456 				  cqe_cnt);
3457 			goto out;
3458 		}
3459 		cqe_cnt -= npolled;
3460 	}
3461 
3462 out:
3463 	mutex_unlock(&free_mr->mutex);
3464 }
3465 
3466 static void hns_roce_v2_dereg_mr(struct hns_roce_dev *hr_dev)
3467 {
3468 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08)
3469 		free_mr_send_cmd_to_hw(hr_dev);
3470 }
3471 
3472 static void *get_cqe_v2(struct hns_roce_cq *hr_cq, int n)
3473 {
3474 	return hns_roce_buf_offset(hr_cq->mtr.kmem, n * hr_cq->cqe_size);
3475 }
3476 
3477 static void *get_sw_cqe_v2(struct hns_roce_cq *hr_cq, unsigned int n)
3478 {
3479 	struct hns_roce_v2_cqe *cqe = get_cqe_v2(hr_cq, n & hr_cq->ib_cq.cqe);
3480 
3481 	/* Get cqe when Owner bit is Conversely with the MSB of cons_idx */
3482 	return (hr_reg_read(cqe, CQE_OWNER) ^ !!(n & hr_cq->cq_depth)) ? cqe :
3483 									 NULL;
3484 }
3485 
3486 static inline void update_cq_db(struct hns_roce_dev *hr_dev,
3487 				struct hns_roce_cq *hr_cq)
3488 {
3489 	if (likely(hr_cq->flags & HNS_ROCE_CQ_FLAG_RECORD_DB)) {
3490 		*hr_cq->set_ci_db = hr_cq->cons_index & V2_CQ_DB_CONS_IDX_M;
3491 	} else {
3492 		struct hns_roce_v2_db cq_db = {};
3493 
3494 		hr_reg_write(&cq_db, DB_TAG, hr_cq->cqn);
3495 		hr_reg_write(&cq_db, DB_CMD, HNS_ROCE_V2_CQ_DB);
3496 		hr_reg_write(&cq_db, DB_CQ_CI, hr_cq->cons_index);
3497 		hr_reg_write(&cq_db, DB_CQ_CMD_SN, 1);
3498 
3499 		hns_roce_write64(hr_dev, (__le32 *)&cq_db, hr_cq->db_reg);
3500 	}
3501 }
3502 
3503 static void __hns_roce_v2_cq_clean(struct hns_roce_cq *hr_cq, u32 qpn,
3504 				   struct hns_roce_srq *srq)
3505 {
3506 	struct hns_roce_dev *hr_dev = to_hr_dev(hr_cq->ib_cq.device);
3507 	struct hns_roce_v2_cqe *cqe, *dest;
3508 	u32 prod_index;
3509 	int nfreed = 0;
3510 	int wqe_index;
3511 	u8 owner_bit;
3512 
3513 	for (prod_index = hr_cq->cons_index; get_sw_cqe_v2(hr_cq, prod_index);
3514 	     ++prod_index) {
3515 		if (prod_index > hr_cq->cons_index + hr_cq->ib_cq.cqe)
3516 			break;
3517 	}
3518 
3519 	/*
3520 	 * Now backwards through the CQ, removing CQ entries
3521 	 * that match our QP by overwriting them with next entries.
3522 	 */
3523 	while ((int) --prod_index - (int) hr_cq->cons_index >= 0) {
3524 		cqe = get_cqe_v2(hr_cq, prod_index & hr_cq->ib_cq.cqe);
3525 		if (hr_reg_read(cqe, CQE_LCL_QPN) == qpn) {
3526 			if (srq && hr_reg_read(cqe, CQE_S_R)) {
3527 				wqe_index = hr_reg_read(cqe, CQE_WQE_IDX);
3528 				hns_roce_free_srq_wqe(srq, wqe_index);
3529 			}
3530 			++nfreed;
3531 		} else if (nfreed) {
3532 			dest = get_cqe_v2(hr_cq, (prod_index + nfreed) &
3533 					  hr_cq->ib_cq.cqe);
3534 			owner_bit = hr_reg_read(dest, CQE_OWNER);
3535 			memcpy(dest, cqe, hr_cq->cqe_size);
3536 			hr_reg_write(dest, CQE_OWNER, owner_bit);
3537 		}
3538 	}
3539 
3540 	if (nfreed) {
3541 		hr_cq->cons_index += nfreed;
3542 		update_cq_db(hr_dev, hr_cq);
3543 	}
3544 }
3545 
3546 static void hns_roce_v2_cq_clean(struct hns_roce_cq *hr_cq, u32 qpn,
3547 				 struct hns_roce_srq *srq)
3548 {
3549 	spin_lock_irq(&hr_cq->lock);
3550 	__hns_roce_v2_cq_clean(hr_cq, qpn, srq);
3551 	spin_unlock_irq(&hr_cq->lock);
3552 }
3553 
3554 static void hns_roce_v2_write_cqc(struct hns_roce_dev *hr_dev,
3555 				  struct hns_roce_cq *hr_cq, void *mb_buf,
3556 				  u64 *mtts, dma_addr_t dma_handle)
3557 {
3558 	struct hns_roce_v2_cq_context *cq_context;
3559 
3560 	cq_context = mb_buf;
3561 	memset(cq_context, 0, sizeof(*cq_context));
3562 
3563 	hr_reg_write(cq_context, CQC_CQ_ST, V2_CQ_STATE_VALID);
3564 	hr_reg_write(cq_context, CQC_ARM_ST, NO_ARMED);
3565 	hr_reg_write(cq_context, CQC_SHIFT, ilog2(hr_cq->cq_depth));
3566 	hr_reg_write(cq_context, CQC_CEQN, hr_cq->vector);
3567 	hr_reg_write(cq_context, CQC_CQN, hr_cq->cqn);
3568 
3569 	if (hr_cq->cqe_size == HNS_ROCE_V3_CQE_SIZE)
3570 		hr_reg_write(cq_context, CQC_CQE_SIZE, CQE_SIZE_64B);
3571 
3572 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_STASH)
3573 		hr_reg_enable(cq_context, CQC_STASH);
3574 
3575 	hr_reg_write(cq_context, CQC_CQE_CUR_BLK_ADDR_L,
3576 		     to_hr_hw_page_addr(mtts[0]));
3577 	hr_reg_write(cq_context, CQC_CQE_CUR_BLK_ADDR_H,
3578 		     upper_32_bits(to_hr_hw_page_addr(mtts[0])));
3579 	hr_reg_write(cq_context, CQC_CQE_HOP_NUM, hr_dev->caps.cqe_hop_num ==
3580 		     HNS_ROCE_HOP_NUM_0 ? 0 : hr_dev->caps.cqe_hop_num);
3581 	hr_reg_write(cq_context, CQC_CQE_NEX_BLK_ADDR_L,
3582 		     to_hr_hw_page_addr(mtts[1]));
3583 	hr_reg_write(cq_context, CQC_CQE_NEX_BLK_ADDR_H,
3584 		     upper_32_bits(to_hr_hw_page_addr(mtts[1])));
3585 	hr_reg_write(cq_context, CQC_CQE_BAR_PG_SZ,
3586 		     to_hr_hw_page_shift(hr_cq->mtr.hem_cfg.ba_pg_shift));
3587 	hr_reg_write(cq_context, CQC_CQE_BUF_PG_SZ,
3588 		     to_hr_hw_page_shift(hr_cq->mtr.hem_cfg.buf_pg_shift));
3589 	hr_reg_write(cq_context, CQC_CQE_BA_L, dma_handle >> 3);
3590 	hr_reg_write(cq_context, CQC_CQE_BA_H, (dma_handle >> (32 + 3)));
3591 	hr_reg_write_bool(cq_context, CQC_DB_RECORD_EN,
3592 			  hr_cq->flags & HNS_ROCE_CQ_FLAG_RECORD_DB);
3593 	hr_reg_write(cq_context, CQC_CQE_DB_RECORD_ADDR_L,
3594 		     ((u32)hr_cq->db.dma) >> 1);
3595 	hr_reg_write(cq_context, CQC_CQE_DB_RECORD_ADDR_H,
3596 		     hr_cq->db.dma >> 32);
3597 	hr_reg_write(cq_context, CQC_CQ_MAX_CNT,
3598 		     HNS_ROCE_V2_CQ_DEFAULT_BURST_NUM);
3599 	hr_reg_write(cq_context, CQC_CQ_PERIOD,
3600 		     HNS_ROCE_V2_CQ_DEFAULT_INTERVAL);
3601 }
3602 
3603 static int hns_roce_v2_req_notify_cq(struct ib_cq *ibcq,
3604 				     enum ib_cq_notify_flags flags)
3605 {
3606 	struct hns_roce_dev *hr_dev = to_hr_dev(ibcq->device);
3607 	struct hns_roce_cq *hr_cq = to_hr_cq(ibcq);
3608 	struct hns_roce_v2_db cq_db = {};
3609 	u32 notify_flag;
3610 
3611 	/*
3612 	 * flags = 0, then notify_flag : next
3613 	 * flags = 1, then notify flag : solocited
3614 	 */
3615 	notify_flag = (flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED ?
3616 		      V2_CQ_DB_REQ_NOT : V2_CQ_DB_REQ_NOT_SOL;
3617 
3618 	hr_reg_write(&cq_db, DB_TAG, hr_cq->cqn);
3619 	hr_reg_write(&cq_db, DB_CMD, HNS_ROCE_V2_CQ_DB_NOTIFY);
3620 	hr_reg_write(&cq_db, DB_CQ_CI, hr_cq->cons_index);
3621 	hr_reg_write(&cq_db, DB_CQ_CMD_SN, hr_cq->arm_sn);
3622 	hr_reg_write(&cq_db, DB_CQ_NOTIFY, notify_flag);
3623 
3624 	hns_roce_write64(hr_dev, (__le32 *)&cq_db, hr_cq->db_reg);
3625 
3626 	return 0;
3627 }
3628 
3629 static int sw_comp(struct hns_roce_qp *hr_qp, struct hns_roce_wq *wq,
3630 		   int num_entries, struct ib_wc *wc)
3631 {
3632 	unsigned int left;
3633 	int npolled = 0;
3634 
3635 	left = wq->head - wq->tail;
3636 	if (left == 0)
3637 		return 0;
3638 
3639 	left = min_t(unsigned int, (unsigned int)num_entries, left);
3640 	while (npolled < left) {
3641 		wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
3642 		wc->status = IB_WC_WR_FLUSH_ERR;
3643 		wc->vendor_err = 0;
3644 		wc->qp = &hr_qp->ibqp;
3645 
3646 		wq->tail++;
3647 		wc++;
3648 		npolled++;
3649 	}
3650 
3651 	return npolled;
3652 }
3653 
3654 static int hns_roce_v2_sw_poll_cq(struct hns_roce_cq *hr_cq, int num_entries,
3655 				  struct ib_wc *wc)
3656 {
3657 	struct hns_roce_qp *hr_qp;
3658 	int npolled = 0;
3659 
3660 	list_for_each_entry(hr_qp, &hr_cq->sq_list, sq_node) {
3661 		npolled += sw_comp(hr_qp, &hr_qp->sq,
3662 				   num_entries - npolled, wc + npolled);
3663 		if (npolled >= num_entries)
3664 			goto out;
3665 	}
3666 
3667 	list_for_each_entry(hr_qp, &hr_cq->rq_list, rq_node) {
3668 		npolled += sw_comp(hr_qp, &hr_qp->rq,
3669 				   num_entries - npolled, wc + npolled);
3670 		if (npolled >= num_entries)
3671 			goto out;
3672 	}
3673 
3674 out:
3675 	return npolled;
3676 }
3677 
3678 static void get_cqe_status(struct hns_roce_dev *hr_dev, struct hns_roce_qp *qp,
3679 			   struct hns_roce_cq *cq, struct hns_roce_v2_cqe *cqe,
3680 			   struct ib_wc *wc)
3681 {
3682 	static const struct {
3683 		u32 cqe_status;
3684 		enum ib_wc_status wc_status;
3685 	} map[] = {
3686 		{ HNS_ROCE_CQE_V2_SUCCESS, IB_WC_SUCCESS },
3687 		{ HNS_ROCE_CQE_V2_LOCAL_LENGTH_ERR, IB_WC_LOC_LEN_ERR },
3688 		{ HNS_ROCE_CQE_V2_LOCAL_QP_OP_ERR, IB_WC_LOC_QP_OP_ERR },
3689 		{ HNS_ROCE_CQE_V2_LOCAL_PROT_ERR, IB_WC_LOC_PROT_ERR },
3690 		{ HNS_ROCE_CQE_V2_WR_FLUSH_ERR, IB_WC_WR_FLUSH_ERR },
3691 		{ HNS_ROCE_CQE_V2_MW_BIND_ERR, IB_WC_MW_BIND_ERR },
3692 		{ HNS_ROCE_CQE_V2_BAD_RESP_ERR, IB_WC_BAD_RESP_ERR },
3693 		{ HNS_ROCE_CQE_V2_LOCAL_ACCESS_ERR, IB_WC_LOC_ACCESS_ERR },
3694 		{ HNS_ROCE_CQE_V2_REMOTE_INVAL_REQ_ERR, IB_WC_REM_INV_REQ_ERR },
3695 		{ HNS_ROCE_CQE_V2_REMOTE_ACCESS_ERR, IB_WC_REM_ACCESS_ERR },
3696 		{ HNS_ROCE_CQE_V2_REMOTE_OP_ERR, IB_WC_REM_OP_ERR },
3697 		{ HNS_ROCE_CQE_V2_TRANSPORT_RETRY_EXC_ERR,
3698 		  IB_WC_RETRY_EXC_ERR },
3699 		{ HNS_ROCE_CQE_V2_RNR_RETRY_EXC_ERR, IB_WC_RNR_RETRY_EXC_ERR },
3700 		{ HNS_ROCE_CQE_V2_REMOTE_ABORT_ERR, IB_WC_REM_ABORT_ERR },
3701 		{ HNS_ROCE_CQE_V2_GENERAL_ERR, IB_WC_GENERAL_ERR}
3702 	};
3703 
3704 	u32 cqe_status = hr_reg_read(cqe, CQE_STATUS);
3705 	int i;
3706 
3707 	wc->status = IB_WC_GENERAL_ERR;
3708 	for (i = 0; i < ARRAY_SIZE(map); i++)
3709 		if (cqe_status == map[i].cqe_status) {
3710 			wc->status = map[i].wc_status;
3711 			break;
3712 		}
3713 
3714 	if (likely(wc->status == IB_WC_SUCCESS ||
3715 		   wc->status == IB_WC_WR_FLUSH_ERR))
3716 		return;
3717 
3718 	ibdev_err(&hr_dev->ib_dev, "error cqe status 0x%x:\n", cqe_status);
3719 	print_hex_dump(KERN_ERR, "", DUMP_PREFIX_NONE, 16, 4, cqe,
3720 		       cq->cqe_size, false);
3721 	wc->vendor_err = hr_reg_read(cqe, CQE_SUB_STATUS);
3722 
3723 	/*
3724 	 * For hns ROCEE, GENERAL_ERR is an error type that is not defined in
3725 	 * the standard protocol, the driver must ignore it and needn't to set
3726 	 * the QP to an error state.
3727 	 */
3728 	if (cqe_status == HNS_ROCE_CQE_V2_GENERAL_ERR)
3729 		return;
3730 
3731 	flush_cqe(hr_dev, qp);
3732 }
3733 
3734 static int get_cur_qp(struct hns_roce_cq *hr_cq, struct hns_roce_v2_cqe *cqe,
3735 		      struct hns_roce_qp **cur_qp)
3736 {
3737 	struct hns_roce_dev *hr_dev = to_hr_dev(hr_cq->ib_cq.device);
3738 	struct hns_roce_qp *hr_qp = *cur_qp;
3739 	u32 qpn;
3740 
3741 	qpn = hr_reg_read(cqe, CQE_LCL_QPN);
3742 
3743 	if (!hr_qp || qpn != hr_qp->qpn) {
3744 		hr_qp = __hns_roce_qp_lookup(hr_dev, qpn);
3745 		if (unlikely(!hr_qp)) {
3746 			ibdev_err(&hr_dev->ib_dev,
3747 				  "CQ %06lx with entry for unknown QPN %06x\n",
3748 				  hr_cq->cqn, qpn);
3749 			return -EINVAL;
3750 		}
3751 		*cur_qp = hr_qp;
3752 	}
3753 
3754 	return 0;
3755 }
3756 
3757 /*
3758  * mapped-value = 1 + real-value
3759  * The ib wc opcode's real value is start from 0, In order to distinguish
3760  * between initialized and uninitialized map values, we plus 1 to the actual
3761  * value when defining the mapping, so that the validity can be identified by
3762  * checking whether the mapped value is greater than 0.
3763  */
3764 #define HR_WC_OP_MAP(hr_key, ib_key) \
3765 		[HNS_ROCE_V2_WQE_OP_ ## hr_key] = 1 + IB_WC_ ## ib_key
3766 
3767 static const u32 wc_send_op_map[] = {
3768 	HR_WC_OP_MAP(SEND,			SEND),
3769 	HR_WC_OP_MAP(SEND_WITH_INV,		SEND),
3770 	HR_WC_OP_MAP(SEND_WITH_IMM,		SEND),
3771 	HR_WC_OP_MAP(RDMA_READ,			RDMA_READ),
3772 	HR_WC_OP_MAP(RDMA_WRITE,		RDMA_WRITE),
3773 	HR_WC_OP_MAP(RDMA_WRITE_WITH_IMM,	RDMA_WRITE),
3774 	HR_WC_OP_MAP(ATOM_CMP_AND_SWAP,		COMP_SWAP),
3775 	HR_WC_OP_MAP(ATOM_FETCH_AND_ADD,	FETCH_ADD),
3776 	HR_WC_OP_MAP(ATOM_MSK_CMP_AND_SWAP,	MASKED_COMP_SWAP),
3777 	HR_WC_OP_MAP(ATOM_MSK_FETCH_AND_ADD,	MASKED_FETCH_ADD),
3778 	HR_WC_OP_MAP(FAST_REG_PMR,		REG_MR),
3779 	HR_WC_OP_MAP(BIND_MW,			REG_MR),
3780 };
3781 
3782 static int to_ib_wc_send_op(u32 hr_opcode)
3783 {
3784 	if (hr_opcode >= ARRAY_SIZE(wc_send_op_map))
3785 		return -EINVAL;
3786 
3787 	return wc_send_op_map[hr_opcode] ? wc_send_op_map[hr_opcode] - 1 :
3788 					   -EINVAL;
3789 }
3790 
3791 static const u32 wc_recv_op_map[] = {
3792 	HR_WC_OP_MAP(RDMA_WRITE_WITH_IMM,		WITH_IMM),
3793 	HR_WC_OP_MAP(SEND,				RECV),
3794 	HR_WC_OP_MAP(SEND_WITH_IMM,			WITH_IMM),
3795 	HR_WC_OP_MAP(SEND_WITH_INV,			RECV),
3796 };
3797 
3798 static int to_ib_wc_recv_op(u32 hr_opcode)
3799 {
3800 	if (hr_opcode >= ARRAY_SIZE(wc_recv_op_map))
3801 		return -EINVAL;
3802 
3803 	return wc_recv_op_map[hr_opcode] ? wc_recv_op_map[hr_opcode] - 1 :
3804 					   -EINVAL;
3805 }
3806 
3807 static void fill_send_wc(struct ib_wc *wc, struct hns_roce_v2_cqe *cqe)
3808 {
3809 	u32 hr_opcode;
3810 	int ib_opcode;
3811 
3812 	wc->wc_flags = 0;
3813 
3814 	hr_opcode = hr_reg_read(cqe, CQE_OPCODE);
3815 	switch (hr_opcode) {
3816 	case HNS_ROCE_V2_WQE_OP_RDMA_READ:
3817 		wc->byte_len = le32_to_cpu(cqe->byte_cnt);
3818 		break;
3819 	case HNS_ROCE_V2_WQE_OP_SEND_WITH_IMM:
3820 	case HNS_ROCE_V2_WQE_OP_RDMA_WRITE_WITH_IMM:
3821 		wc->wc_flags |= IB_WC_WITH_IMM;
3822 		break;
3823 	case HNS_ROCE_V2_WQE_OP_ATOM_CMP_AND_SWAP:
3824 	case HNS_ROCE_V2_WQE_OP_ATOM_FETCH_AND_ADD:
3825 	case HNS_ROCE_V2_WQE_OP_ATOM_MSK_CMP_AND_SWAP:
3826 	case HNS_ROCE_V2_WQE_OP_ATOM_MSK_FETCH_AND_ADD:
3827 		wc->byte_len  = 8;
3828 		break;
3829 	default:
3830 		break;
3831 	}
3832 
3833 	ib_opcode = to_ib_wc_send_op(hr_opcode);
3834 	if (ib_opcode < 0)
3835 		wc->status = IB_WC_GENERAL_ERR;
3836 	else
3837 		wc->opcode = ib_opcode;
3838 }
3839 
3840 static int fill_recv_wc(struct ib_wc *wc, struct hns_roce_v2_cqe *cqe)
3841 {
3842 	u32 hr_opcode;
3843 	int ib_opcode;
3844 
3845 	wc->byte_len = le32_to_cpu(cqe->byte_cnt);
3846 
3847 	hr_opcode = hr_reg_read(cqe, CQE_OPCODE);
3848 	switch (hr_opcode) {
3849 	case HNS_ROCE_V2_OPCODE_RDMA_WRITE_IMM:
3850 	case HNS_ROCE_V2_OPCODE_SEND_WITH_IMM:
3851 		wc->wc_flags = IB_WC_WITH_IMM;
3852 		wc->ex.imm_data = cpu_to_be32(le32_to_cpu(cqe->immtdata));
3853 		break;
3854 	case HNS_ROCE_V2_OPCODE_SEND_WITH_INV:
3855 		wc->wc_flags = IB_WC_WITH_INVALIDATE;
3856 		wc->ex.invalidate_rkey = le32_to_cpu(cqe->rkey);
3857 		break;
3858 	default:
3859 		wc->wc_flags = 0;
3860 	}
3861 
3862 	ib_opcode = to_ib_wc_recv_op(hr_opcode);
3863 	if (ib_opcode < 0)
3864 		wc->status = IB_WC_GENERAL_ERR;
3865 	else
3866 		wc->opcode = ib_opcode;
3867 
3868 	wc->sl = hr_reg_read(cqe, CQE_SL);
3869 	wc->src_qp = hr_reg_read(cqe, CQE_RMT_QPN);
3870 	wc->slid = 0;
3871 	wc->wc_flags |= hr_reg_read(cqe, CQE_GRH) ? IB_WC_GRH : 0;
3872 	wc->port_num = hr_reg_read(cqe, CQE_PORTN);
3873 	wc->pkey_index = 0;
3874 
3875 	if (hr_reg_read(cqe, CQE_VID_VLD)) {
3876 		wc->vlan_id = hr_reg_read(cqe, CQE_VID);
3877 		wc->wc_flags |= IB_WC_WITH_VLAN;
3878 	} else {
3879 		wc->vlan_id = 0xffff;
3880 	}
3881 
3882 	wc->network_hdr_type = hr_reg_read(cqe, CQE_PORT_TYPE);
3883 
3884 	return 0;
3885 }
3886 
3887 static int hns_roce_v2_poll_one(struct hns_roce_cq *hr_cq,
3888 				struct hns_roce_qp **cur_qp, struct ib_wc *wc)
3889 {
3890 	struct hns_roce_dev *hr_dev = to_hr_dev(hr_cq->ib_cq.device);
3891 	struct hns_roce_qp *qp = *cur_qp;
3892 	struct hns_roce_srq *srq = NULL;
3893 	struct hns_roce_v2_cqe *cqe;
3894 	struct hns_roce_wq *wq;
3895 	int is_send;
3896 	u16 wqe_idx;
3897 	int ret;
3898 
3899 	cqe = get_sw_cqe_v2(hr_cq, hr_cq->cons_index);
3900 	if (!cqe)
3901 		return -EAGAIN;
3902 
3903 	++hr_cq->cons_index;
3904 	/* Memory barrier */
3905 	rmb();
3906 
3907 	ret = get_cur_qp(hr_cq, cqe, &qp);
3908 	if (ret)
3909 		return ret;
3910 
3911 	wc->qp = &qp->ibqp;
3912 	wc->vendor_err = 0;
3913 
3914 	wqe_idx = hr_reg_read(cqe, CQE_WQE_IDX);
3915 
3916 	is_send = !hr_reg_read(cqe, CQE_S_R);
3917 	if (is_send) {
3918 		wq = &qp->sq;
3919 
3920 		/* If sg_signal_bit is set, tail pointer will be updated to
3921 		 * the WQE corresponding to the current CQE.
3922 		 */
3923 		if (qp->sq_signal_bits)
3924 			wq->tail += (wqe_idx - (u16)wq->tail) &
3925 				    (wq->wqe_cnt - 1);
3926 
3927 		wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
3928 		++wq->tail;
3929 
3930 		fill_send_wc(wc, cqe);
3931 	} else {
3932 		if (qp->ibqp.srq) {
3933 			srq = to_hr_srq(qp->ibqp.srq);
3934 			wc->wr_id = srq->wrid[wqe_idx];
3935 			hns_roce_free_srq_wqe(srq, wqe_idx);
3936 		} else {
3937 			wq = &qp->rq;
3938 			wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
3939 			++wq->tail;
3940 		}
3941 
3942 		ret = fill_recv_wc(wc, cqe);
3943 	}
3944 
3945 	get_cqe_status(hr_dev, qp, hr_cq, cqe, wc);
3946 	if (unlikely(wc->status != IB_WC_SUCCESS))
3947 		return 0;
3948 
3949 	return ret;
3950 }
3951 
3952 static int hns_roce_v2_poll_cq(struct ib_cq *ibcq, int num_entries,
3953 			       struct ib_wc *wc)
3954 {
3955 	struct hns_roce_dev *hr_dev = to_hr_dev(ibcq->device);
3956 	struct hns_roce_cq *hr_cq = to_hr_cq(ibcq);
3957 	struct hns_roce_qp *cur_qp = NULL;
3958 	unsigned long flags;
3959 	int npolled;
3960 
3961 	spin_lock_irqsave(&hr_cq->lock, flags);
3962 
3963 	/*
3964 	 * When the device starts to reset, the state is RST_DOWN. At this time,
3965 	 * there may still be some valid CQEs in the hardware that are not
3966 	 * polled. Therefore, it is not allowed to switch to the software mode
3967 	 * immediately. When the state changes to UNINIT, CQE no longer exists
3968 	 * in the hardware, and then switch to software mode.
3969 	 */
3970 	if (hr_dev->state == HNS_ROCE_DEVICE_STATE_UNINIT) {
3971 		npolled = hns_roce_v2_sw_poll_cq(hr_cq, num_entries, wc);
3972 		goto out;
3973 	}
3974 
3975 	for (npolled = 0; npolled < num_entries; ++npolled) {
3976 		if (hns_roce_v2_poll_one(hr_cq, &cur_qp, wc + npolled))
3977 			break;
3978 	}
3979 
3980 	if (npolled)
3981 		update_cq_db(hr_dev, hr_cq);
3982 
3983 out:
3984 	spin_unlock_irqrestore(&hr_cq->lock, flags);
3985 
3986 	return npolled;
3987 }
3988 
3989 static int get_op_for_set_hem(struct hns_roce_dev *hr_dev, u32 type,
3990 			      u32 step_idx, u8 *mbox_cmd)
3991 {
3992 	u8 cmd;
3993 
3994 	switch (type) {
3995 	case HEM_TYPE_QPC:
3996 		cmd = HNS_ROCE_CMD_WRITE_QPC_BT0;
3997 		break;
3998 	case HEM_TYPE_MTPT:
3999 		cmd = HNS_ROCE_CMD_WRITE_MPT_BT0;
4000 		break;
4001 	case HEM_TYPE_CQC:
4002 		cmd = HNS_ROCE_CMD_WRITE_CQC_BT0;
4003 		break;
4004 	case HEM_TYPE_SRQC:
4005 		cmd = HNS_ROCE_CMD_WRITE_SRQC_BT0;
4006 		break;
4007 	case HEM_TYPE_SCCC:
4008 		cmd = HNS_ROCE_CMD_WRITE_SCCC_BT0;
4009 		break;
4010 	case HEM_TYPE_QPC_TIMER:
4011 		cmd = HNS_ROCE_CMD_WRITE_QPC_TIMER_BT0;
4012 		break;
4013 	case HEM_TYPE_CQC_TIMER:
4014 		cmd = HNS_ROCE_CMD_WRITE_CQC_TIMER_BT0;
4015 		break;
4016 	default:
4017 		dev_warn(hr_dev->dev, "failed to check hem type %u.\n", type);
4018 		return -EINVAL;
4019 	}
4020 
4021 	*mbox_cmd = cmd + step_idx;
4022 
4023 	return 0;
4024 }
4025 
4026 static int config_gmv_ba_to_hw(struct hns_roce_dev *hr_dev, unsigned long obj,
4027 			       dma_addr_t base_addr)
4028 {
4029 	struct hns_roce_cmq_desc desc;
4030 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
4031 	u32 idx = obj / (HNS_HW_PAGE_SIZE / hr_dev->caps.gmv_entry_sz);
4032 	u64 addr = to_hr_hw_page_addr(base_addr);
4033 
4034 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_GMV_BT, false);
4035 
4036 	hr_reg_write(req, CFG_GMV_BT_BA_L, lower_32_bits(addr));
4037 	hr_reg_write(req, CFG_GMV_BT_BA_H, upper_32_bits(addr));
4038 	hr_reg_write(req, CFG_GMV_BT_IDX, idx);
4039 
4040 	return hns_roce_cmq_send(hr_dev, &desc, 1);
4041 }
4042 
4043 static int set_hem_to_hw(struct hns_roce_dev *hr_dev, int obj,
4044 			 dma_addr_t base_addr, u32 hem_type, u32 step_idx)
4045 {
4046 	int ret;
4047 	u8 cmd;
4048 
4049 	if (unlikely(hem_type == HEM_TYPE_GMV))
4050 		return config_gmv_ba_to_hw(hr_dev, obj, base_addr);
4051 
4052 	if (unlikely(hem_type == HEM_TYPE_SCCC && step_idx))
4053 		return 0;
4054 
4055 	ret = get_op_for_set_hem(hr_dev, hem_type, step_idx, &cmd);
4056 	if (ret < 0)
4057 		return ret;
4058 
4059 	return config_hem_ba_to_hw(hr_dev, base_addr, cmd, obj);
4060 }
4061 
4062 static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev,
4063 			       struct hns_roce_hem_table *table, int obj,
4064 			       u32 step_idx)
4065 {
4066 	struct hns_roce_hem_iter iter;
4067 	struct hns_roce_hem_mhop mhop;
4068 	struct hns_roce_hem *hem;
4069 	unsigned long mhop_obj = obj;
4070 	int i, j, k;
4071 	int ret = 0;
4072 	u64 hem_idx = 0;
4073 	u64 l1_idx = 0;
4074 	u64 bt_ba = 0;
4075 	u32 chunk_ba_num;
4076 	u32 hop_num;
4077 
4078 	if (!hns_roce_check_whether_mhop(hr_dev, table->type))
4079 		return 0;
4080 
4081 	hns_roce_calc_hem_mhop(hr_dev, table, &mhop_obj, &mhop);
4082 	i = mhop.l0_idx;
4083 	j = mhop.l1_idx;
4084 	k = mhop.l2_idx;
4085 	hop_num = mhop.hop_num;
4086 	chunk_ba_num = mhop.bt_chunk_size / 8;
4087 
4088 	if (hop_num == 2) {
4089 		hem_idx = i * chunk_ba_num * chunk_ba_num + j * chunk_ba_num +
4090 			  k;
4091 		l1_idx = i * chunk_ba_num + j;
4092 	} else if (hop_num == 1) {
4093 		hem_idx = i * chunk_ba_num + j;
4094 	} else if (hop_num == HNS_ROCE_HOP_NUM_0) {
4095 		hem_idx = i;
4096 	}
4097 
4098 	if (table->type == HEM_TYPE_SCCC)
4099 		obj = mhop.l0_idx;
4100 
4101 	if (check_whether_last_step(hop_num, step_idx)) {
4102 		hem = table->hem[hem_idx];
4103 		for (hns_roce_hem_first(hem, &iter);
4104 		     !hns_roce_hem_last(&iter); hns_roce_hem_next(&iter)) {
4105 			bt_ba = hns_roce_hem_addr(&iter);
4106 			ret = set_hem_to_hw(hr_dev, obj, bt_ba, table->type,
4107 					    step_idx);
4108 		}
4109 	} else {
4110 		if (step_idx == 0)
4111 			bt_ba = table->bt_l0_dma_addr[i];
4112 		else if (step_idx == 1 && hop_num == 2)
4113 			bt_ba = table->bt_l1_dma_addr[l1_idx];
4114 
4115 		ret = set_hem_to_hw(hr_dev, obj, bt_ba, table->type, step_idx);
4116 	}
4117 
4118 	return ret;
4119 }
4120 
4121 static int hns_roce_v2_clear_hem(struct hns_roce_dev *hr_dev,
4122 				 struct hns_roce_hem_table *table,
4123 				 int tag, u32 step_idx)
4124 {
4125 	struct hns_roce_cmd_mailbox *mailbox;
4126 	struct device *dev = hr_dev->dev;
4127 	u8 cmd = 0xff;
4128 	int ret;
4129 
4130 	if (!hns_roce_check_whether_mhop(hr_dev, table->type))
4131 		return 0;
4132 
4133 	switch (table->type) {
4134 	case HEM_TYPE_QPC:
4135 		cmd = HNS_ROCE_CMD_DESTROY_QPC_BT0;
4136 		break;
4137 	case HEM_TYPE_MTPT:
4138 		cmd = HNS_ROCE_CMD_DESTROY_MPT_BT0;
4139 		break;
4140 	case HEM_TYPE_CQC:
4141 		cmd = HNS_ROCE_CMD_DESTROY_CQC_BT0;
4142 		break;
4143 	case HEM_TYPE_SRQC:
4144 		cmd = HNS_ROCE_CMD_DESTROY_SRQC_BT0;
4145 		break;
4146 	case HEM_TYPE_SCCC:
4147 	case HEM_TYPE_QPC_TIMER:
4148 	case HEM_TYPE_CQC_TIMER:
4149 	case HEM_TYPE_GMV:
4150 		return 0;
4151 	default:
4152 		dev_warn(dev, "table %u not to be destroyed by mailbox!\n",
4153 			 table->type);
4154 		return 0;
4155 	}
4156 
4157 	cmd += step_idx;
4158 
4159 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
4160 	if (IS_ERR(mailbox))
4161 		return PTR_ERR(mailbox);
4162 
4163 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, cmd, tag);
4164 
4165 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
4166 	return ret;
4167 }
4168 
4169 static int hns_roce_v2_qp_modify(struct hns_roce_dev *hr_dev,
4170 				 struct hns_roce_v2_qp_context *context,
4171 				 struct hns_roce_v2_qp_context *qpc_mask,
4172 				 struct hns_roce_qp *hr_qp)
4173 {
4174 	struct hns_roce_cmd_mailbox *mailbox;
4175 	int qpc_size;
4176 	int ret;
4177 
4178 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
4179 	if (IS_ERR(mailbox))
4180 		return PTR_ERR(mailbox);
4181 
4182 	/* The qpc size of HIP08 is only 256B, which is half of HIP09 */
4183 	qpc_size = hr_dev->caps.qpc_sz;
4184 	memcpy(mailbox->buf, context, qpc_size);
4185 	memcpy(mailbox->buf + qpc_size, qpc_mask, qpc_size);
4186 
4187 	ret = hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0,
4188 				HNS_ROCE_CMD_MODIFY_QPC, hr_qp->qpn);
4189 
4190 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
4191 
4192 	return ret;
4193 }
4194 
4195 static void set_access_flags(struct hns_roce_qp *hr_qp,
4196 			     struct hns_roce_v2_qp_context *context,
4197 			     struct hns_roce_v2_qp_context *qpc_mask,
4198 			     const struct ib_qp_attr *attr, int attr_mask)
4199 {
4200 	u8 dest_rd_atomic;
4201 	u32 access_flags;
4202 
4203 	dest_rd_atomic = (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) ?
4204 			 attr->max_dest_rd_atomic : hr_qp->resp_depth;
4205 
4206 	access_flags = (attr_mask & IB_QP_ACCESS_FLAGS) ?
4207 		       attr->qp_access_flags : hr_qp->atomic_rd_en;
4208 
4209 	if (!dest_rd_atomic)
4210 		access_flags &= IB_ACCESS_REMOTE_WRITE;
4211 
4212 	hr_reg_write_bool(context, QPC_RRE,
4213 			  access_flags & IB_ACCESS_REMOTE_READ);
4214 	hr_reg_clear(qpc_mask, QPC_RRE);
4215 
4216 	hr_reg_write_bool(context, QPC_RWE,
4217 			  access_flags & IB_ACCESS_REMOTE_WRITE);
4218 	hr_reg_clear(qpc_mask, QPC_RWE);
4219 
4220 	hr_reg_write_bool(context, QPC_ATE,
4221 			  access_flags & IB_ACCESS_REMOTE_ATOMIC);
4222 	hr_reg_clear(qpc_mask, QPC_ATE);
4223 	hr_reg_write_bool(context, QPC_EXT_ATE,
4224 			  access_flags & IB_ACCESS_REMOTE_ATOMIC);
4225 	hr_reg_clear(qpc_mask, QPC_EXT_ATE);
4226 }
4227 
4228 static void set_qpc_wqe_cnt(struct hns_roce_qp *hr_qp,
4229 			    struct hns_roce_v2_qp_context *context,
4230 			    struct hns_roce_v2_qp_context *qpc_mask)
4231 {
4232 	hr_reg_write(context, QPC_SGE_SHIFT,
4233 		     to_hr_hem_entries_shift(hr_qp->sge.sge_cnt,
4234 					     hr_qp->sge.sge_shift));
4235 
4236 	hr_reg_write(context, QPC_SQ_SHIFT, ilog2(hr_qp->sq.wqe_cnt));
4237 
4238 	hr_reg_write(context, QPC_RQ_SHIFT, ilog2(hr_qp->rq.wqe_cnt));
4239 }
4240 
4241 static inline int get_cqn(struct ib_cq *ib_cq)
4242 {
4243 	return ib_cq ? to_hr_cq(ib_cq)->cqn : 0;
4244 }
4245 
4246 static inline int get_pdn(struct ib_pd *ib_pd)
4247 {
4248 	return ib_pd ? to_hr_pd(ib_pd)->pdn : 0;
4249 }
4250 
4251 static void modify_qp_reset_to_init(struct ib_qp *ibqp,
4252 				    const struct ib_qp_attr *attr,
4253 				    struct hns_roce_v2_qp_context *context,
4254 				    struct hns_roce_v2_qp_context *qpc_mask)
4255 {
4256 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4257 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4258 
4259 	/*
4260 	 * In v2 engine, software pass context and context mask to hardware
4261 	 * when modifying qp. If software need modify some fields in context,
4262 	 * we should set all bits of the relevant fields in context mask to
4263 	 * 0 at the same time, else set them to 0x1.
4264 	 */
4265 	hr_reg_write(context, QPC_TST, to_hr_qp_type(ibqp->qp_type));
4266 
4267 	hr_reg_write(context, QPC_PD, get_pdn(ibqp->pd));
4268 
4269 	hr_reg_write(context, QPC_RQWS, ilog2(hr_qp->rq.max_gs));
4270 
4271 	set_qpc_wqe_cnt(hr_qp, context, qpc_mask);
4272 
4273 	/* No VLAN need to set 0xFFF */
4274 	hr_reg_write(context, QPC_VLAN_ID, 0xfff);
4275 
4276 	if (ibqp->qp_type == IB_QPT_XRC_TGT) {
4277 		context->qkey_xrcd = cpu_to_le32(hr_qp->xrcdn);
4278 
4279 		hr_reg_enable(context, QPC_XRC_QP_TYPE);
4280 	}
4281 
4282 	if (hr_qp->en_flags & HNS_ROCE_QP_CAP_RQ_RECORD_DB)
4283 		hr_reg_enable(context, QPC_RQ_RECORD_EN);
4284 
4285 	if (hr_qp->en_flags & HNS_ROCE_QP_CAP_OWNER_DB)
4286 		hr_reg_enable(context, QPC_OWNER_MODE);
4287 
4288 	hr_reg_write(context, QPC_RQ_DB_RECORD_ADDR_L,
4289 		     lower_32_bits(hr_qp->rdb.dma) >> 1);
4290 	hr_reg_write(context, QPC_RQ_DB_RECORD_ADDR_H,
4291 		     upper_32_bits(hr_qp->rdb.dma));
4292 
4293 	hr_reg_write(context, QPC_RX_CQN, get_cqn(ibqp->recv_cq));
4294 
4295 	if (ibqp->srq) {
4296 		hr_reg_enable(context, QPC_SRQ_EN);
4297 		hr_reg_write(context, QPC_SRQN, to_hr_srq(ibqp->srq)->srqn);
4298 	}
4299 
4300 	hr_reg_enable(context, QPC_FRE);
4301 
4302 	hr_reg_write(context, QPC_TX_CQN, get_cqn(ibqp->send_cq));
4303 
4304 	if (hr_dev->caps.qpc_sz < HNS_ROCE_V3_QPC_SZ)
4305 		return;
4306 
4307 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_STASH)
4308 		hr_reg_enable(&context->ext, QPCEX_STASH);
4309 }
4310 
4311 static void modify_qp_init_to_init(struct ib_qp *ibqp,
4312 				   const struct ib_qp_attr *attr,
4313 				   struct hns_roce_v2_qp_context *context,
4314 				   struct hns_roce_v2_qp_context *qpc_mask)
4315 {
4316 	/*
4317 	 * In v2 engine, software pass context and context mask to hardware
4318 	 * when modifying qp. If software need modify some fields in context,
4319 	 * we should set all bits of the relevant fields in context mask to
4320 	 * 0 at the same time, else set them to 0x1.
4321 	 */
4322 	hr_reg_write(context, QPC_TST, to_hr_qp_type(ibqp->qp_type));
4323 	hr_reg_clear(qpc_mask, QPC_TST);
4324 
4325 	hr_reg_write(context, QPC_PD, get_pdn(ibqp->pd));
4326 	hr_reg_clear(qpc_mask, QPC_PD);
4327 
4328 	hr_reg_write(context, QPC_RX_CQN, get_cqn(ibqp->recv_cq));
4329 	hr_reg_clear(qpc_mask, QPC_RX_CQN);
4330 
4331 	hr_reg_write(context, QPC_TX_CQN, get_cqn(ibqp->send_cq));
4332 	hr_reg_clear(qpc_mask, QPC_TX_CQN);
4333 
4334 	if (ibqp->srq) {
4335 		hr_reg_enable(context, QPC_SRQ_EN);
4336 		hr_reg_clear(qpc_mask, QPC_SRQ_EN);
4337 		hr_reg_write(context, QPC_SRQN, to_hr_srq(ibqp->srq)->srqn);
4338 		hr_reg_clear(qpc_mask, QPC_SRQN);
4339 	}
4340 }
4341 
4342 static int config_qp_rq_buf(struct hns_roce_dev *hr_dev,
4343 			    struct hns_roce_qp *hr_qp,
4344 			    struct hns_roce_v2_qp_context *context,
4345 			    struct hns_roce_v2_qp_context *qpc_mask)
4346 {
4347 	u64 mtts[MTT_MIN_COUNT] = { 0 };
4348 	u64 wqe_sge_ba;
4349 	int count;
4350 
4351 	/* Search qp buf's mtts */
4352 	count = hns_roce_mtr_find(hr_dev, &hr_qp->mtr, hr_qp->rq.offset, mtts,
4353 				  MTT_MIN_COUNT, &wqe_sge_ba);
4354 	if (hr_qp->rq.wqe_cnt && count < 1) {
4355 		ibdev_err(&hr_dev->ib_dev,
4356 			  "failed to find RQ WQE, QPN = 0x%lx.\n", hr_qp->qpn);
4357 		return -EINVAL;
4358 	}
4359 
4360 	context->wqe_sge_ba = cpu_to_le32(wqe_sge_ba >> 3);
4361 	qpc_mask->wqe_sge_ba = 0;
4362 
4363 	/*
4364 	 * In v2 engine, software pass context and context mask to hardware
4365 	 * when modifying qp. If software need modify some fields in context,
4366 	 * we should set all bits of the relevant fields in context mask to
4367 	 * 0 at the same time, else set them to 0x1.
4368 	 */
4369 	hr_reg_write(context, QPC_WQE_SGE_BA_H, wqe_sge_ba >> (32 + 3));
4370 	hr_reg_clear(qpc_mask, QPC_WQE_SGE_BA_H);
4371 
4372 	hr_reg_write(context, QPC_SQ_HOP_NUM,
4373 		     to_hr_hem_hopnum(hr_dev->caps.wqe_sq_hop_num,
4374 				      hr_qp->sq.wqe_cnt));
4375 	hr_reg_clear(qpc_mask, QPC_SQ_HOP_NUM);
4376 
4377 	hr_reg_write(context, QPC_SGE_HOP_NUM,
4378 		     to_hr_hem_hopnum(hr_dev->caps.wqe_sge_hop_num,
4379 				      hr_qp->sge.sge_cnt));
4380 	hr_reg_clear(qpc_mask, QPC_SGE_HOP_NUM);
4381 
4382 	hr_reg_write(context, QPC_RQ_HOP_NUM,
4383 		     to_hr_hem_hopnum(hr_dev->caps.wqe_rq_hop_num,
4384 				      hr_qp->rq.wqe_cnt));
4385 
4386 	hr_reg_clear(qpc_mask, QPC_RQ_HOP_NUM);
4387 
4388 	hr_reg_write(context, QPC_WQE_SGE_BA_PG_SZ,
4389 		     to_hr_hw_page_shift(hr_qp->mtr.hem_cfg.ba_pg_shift));
4390 	hr_reg_clear(qpc_mask, QPC_WQE_SGE_BA_PG_SZ);
4391 
4392 	hr_reg_write(context, QPC_WQE_SGE_BUF_PG_SZ,
4393 		     to_hr_hw_page_shift(hr_qp->mtr.hem_cfg.buf_pg_shift));
4394 	hr_reg_clear(qpc_mask, QPC_WQE_SGE_BUF_PG_SZ);
4395 
4396 	context->rq_cur_blk_addr = cpu_to_le32(to_hr_hw_page_addr(mtts[0]));
4397 	qpc_mask->rq_cur_blk_addr = 0;
4398 
4399 	hr_reg_write(context, QPC_RQ_CUR_BLK_ADDR_H,
4400 		     upper_32_bits(to_hr_hw_page_addr(mtts[0])));
4401 	hr_reg_clear(qpc_mask, QPC_RQ_CUR_BLK_ADDR_H);
4402 
4403 	context->rq_nxt_blk_addr = cpu_to_le32(to_hr_hw_page_addr(mtts[1]));
4404 	qpc_mask->rq_nxt_blk_addr = 0;
4405 
4406 	hr_reg_write(context, QPC_RQ_NXT_BLK_ADDR_H,
4407 		     upper_32_bits(to_hr_hw_page_addr(mtts[1])));
4408 	hr_reg_clear(qpc_mask, QPC_RQ_NXT_BLK_ADDR_H);
4409 
4410 	return 0;
4411 }
4412 
4413 static int config_qp_sq_buf(struct hns_roce_dev *hr_dev,
4414 			    struct hns_roce_qp *hr_qp,
4415 			    struct hns_roce_v2_qp_context *context,
4416 			    struct hns_roce_v2_qp_context *qpc_mask)
4417 {
4418 	struct ib_device *ibdev = &hr_dev->ib_dev;
4419 	u64 sge_cur_blk = 0;
4420 	u64 sq_cur_blk = 0;
4421 	int count;
4422 
4423 	/* search qp buf's mtts */
4424 	count = hns_roce_mtr_find(hr_dev, &hr_qp->mtr, 0, &sq_cur_blk, 1, NULL);
4425 	if (count < 1) {
4426 		ibdev_err(ibdev, "failed to find QP(0x%lx) SQ buf.\n",
4427 			  hr_qp->qpn);
4428 		return -EINVAL;
4429 	}
4430 	if (hr_qp->sge.sge_cnt > 0) {
4431 		count = hns_roce_mtr_find(hr_dev, &hr_qp->mtr,
4432 					  hr_qp->sge.offset,
4433 					  &sge_cur_blk, 1, NULL);
4434 		if (count < 1) {
4435 			ibdev_err(ibdev, "failed to find QP(0x%lx) SGE buf.\n",
4436 				  hr_qp->qpn);
4437 			return -EINVAL;
4438 		}
4439 	}
4440 
4441 	/*
4442 	 * In v2 engine, software pass context and context mask to hardware
4443 	 * when modifying qp. If software need modify some fields in context,
4444 	 * we should set all bits of the relevant fields in context mask to
4445 	 * 0 at the same time, else set them to 0x1.
4446 	 */
4447 	hr_reg_write(context, QPC_SQ_CUR_BLK_ADDR_L,
4448 		     lower_32_bits(to_hr_hw_page_addr(sq_cur_blk)));
4449 	hr_reg_write(context, QPC_SQ_CUR_BLK_ADDR_H,
4450 		     upper_32_bits(to_hr_hw_page_addr(sq_cur_blk)));
4451 	hr_reg_clear(qpc_mask, QPC_SQ_CUR_BLK_ADDR_L);
4452 	hr_reg_clear(qpc_mask, QPC_SQ_CUR_BLK_ADDR_H);
4453 
4454 	hr_reg_write(context, QPC_SQ_CUR_SGE_BLK_ADDR_L,
4455 		     lower_32_bits(to_hr_hw_page_addr(sge_cur_blk)));
4456 	hr_reg_write(context, QPC_SQ_CUR_SGE_BLK_ADDR_H,
4457 		     upper_32_bits(to_hr_hw_page_addr(sge_cur_blk)));
4458 	hr_reg_clear(qpc_mask, QPC_SQ_CUR_SGE_BLK_ADDR_L);
4459 	hr_reg_clear(qpc_mask, QPC_SQ_CUR_SGE_BLK_ADDR_H);
4460 
4461 	hr_reg_write(context, QPC_RX_SQ_CUR_BLK_ADDR_L,
4462 		     lower_32_bits(to_hr_hw_page_addr(sq_cur_blk)));
4463 	hr_reg_write(context, QPC_RX_SQ_CUR_BLK_ADDR_H,
4464 		     upper_32_bits(to_hr_hw_page_addr(sq_cur_blk)));
4465 	hr_reg_clear(qpc_mask, QPC_RX_SQ_CUR_BLK_ADDR_L);
4466 	hr_reg_clear(qpc_mask, QPC_RX_SQ_CUR_BLK_ADDR_H);
4467 
4468 	return 0;
4469 }
4470 
4471 static inline enum ib_mtu get_mtu(struct ib_qp *ibqp,
4472 				  const struct ib_qp_attr *attr)
4473 {
4474 	if (ibqp->qp_type == IB_QPT_GSI || ibqp->qp_type == IB_QPT_UD)
4475 		return IB_MTU_4096;
4476 
4477 	return attr->path_mtu;
4478 }
4479 
4480 static int modify_qp_init_to_rtr(struct ib_qp *ibqp,
4481 				 const struct ib_qp_attr *attr, int attr_mask,
4482 				 struct hns_roce_v2_qp_context *context,
4483 				 struct hns_roce_v2_qp_context *qpc_mask,
4484 				 struct ib_udata *udata)
4485 {
4486 	struct hns_roce_ucontext *uctx = rdma_udata_to_drv_context(udata,
4487 					  struct hns_roce_ucontext, ibucontext);
4488 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4489 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4490 	struct ib_device *ibdev = &hr_dev->ib_dev;
4491 	dma_addr_t trrl_ba;
4492 	dma_addr_t irrl_ba;
4493 	enum ib_mtu ib_mtu;
4494 	const u8 *smac;
4495 	u8 lp_pktn_ini;
4496 	u64 *mtts;
4497 	u8 *dmac;
4498 	u32 port;
4499 	int mtu;
4500 	int ret;
4501 
4502 	ret = config_qp_rq_buf(hr_dev, hr_qp, context, qpc_mask);
4503 	if (ret) {
4504 		ibdev_err(ibdev, "failed to config rq buf, ret = %d.\n", ret);
4505 		return ret;
4506 	}
4507 
4508 	/* Search IRRL's mtts */
4509 	mtts = hns_roce_table_find(hr_dev, &hr_dev->qp_table.irrl_table,
4510 				   hr_qp->qpn, &irrl_ba);
4511 	if (!mtts) {
4512 		ibdev_err(ibdev, "failed to find qp irrl_table.\n");
4513 		return -EINVAL;
4514 	}
4515 
4516 	/* Search TRRL's mtts */
4517 	mtts = hns_roce_table_find(hr_dev, &hr_dev->qp_table.trrl_table,
4518 				   hr_qp->qpn, &trrl_ba);
4519 	if (!mtts) {
4520 		ibdev_err(ibdev, "failed to find qp trrl_table.\n");
4521 		return -EINVAL;
4522 	}
4523 
4524 	if (attr_mask & IB_QP_ALT_PATH) {
4525 		ibdev_err(ibdev, "INIT2RTR attr_mask (0x%x) error.\n",
4526 			  attr_mask);
4527 		return -EINVAL;
4528 	}
4529 
4530 	hr_reg_write(context, QPC_TRRL_BA_L, trrl_ba >> 4);
4531 	hr_reg_clear(qpc_mask, QPC_TRRL_BA_L);
4532 	context->trrl_ba = cpu_to_le32(trrl_ba >> (16 + 4));
4533 	qpc_mask->trrl_ba = 0;
4534 	hr_reg_write(context, QPC_TRRL_BA_H, trrl_ba >> (32 + 16 + 4));
4535 	hr_reg_clear(qpc_mask, QPC_TRRL_BA_H);
4536 
4537 	context->irrl_ba = cpu_to_le32(irrl_ba >> 6);
4538 	qpc_mask->irrl_ba = 0;
4539 	hr_reg_write(context, QPC_IRRL_BA_H, irrl_ba >> (32 + 6));
4540 	hr_reg_clear(qpc_mask, QPC_IRRL_BA_H);
4541 
4542 	hr_reg_enable(context, QPC_RMT_E2E);
4543 	hr_reg_clear(qpc_mask, QPC_RMT_E2E);
4544 
4545 	hr_reg_write(context, QPC_SIG_TYPE, hr_qp->sq_signal_bits);
4546 	hr_reg_clear(qpc_mask, QPC_SIG_TYPE);
4547 
4548 	port = (attr_mask & IB_QP_PORT) ? (attr->port_num - 1) : hr_qp->port;
4549 
4550 	smac = (const u8 *)hr_dev->dev_addr[port];
4551 	dmac = (u8 *)attr->ah_attr.roce.dmac;
4552 	/* when dmac equals smac or loop_idc is 1, it should loopback */
4553 	if (ether_addr_equal_unaligned(dmac, smac) ||
4554 	    hr_dev->loop_idc == 0x1) {
4555 		hr_reg_write(context, QPC_LBI, hr_dev->loop_idc);
4556 		hr_reg_clear(qpc_mask, QPC_LBI);
4557 	}
4558 
4559 	if (attr_mask & IB_QP_DEST_QPN) {
4560 		hr_reg_write(context, QPC_DQPN, attr->dest_qp_num);
4561 		hr_reg_clear(qpc_mask, QPC_DQPN);
4562 	}
4563 
4564 	memcpy(&context->dmac, dmac, sizeof(u32));
4565 	hr_reg_write(context, QPC_DMAC_H, *((u16 *)(&dmac[4])));
4566 	qpc_mask->dmac = 0;
4567 	hr_reg_clear(qpc_mask, QPC_DMAC_H);
4568 
4569 	ib_mtu = get_mtu(ibqp, attr);
4570 	hr_qp->path_mtu = ib_mtu;
4571 
4572 	mtu = ib_mtu_enum_to_int(ib_mtu);
4573 	if (WARN_ON(mtu <= 0))
4574 		return -EINVAL;
4575 #define MIN_LP_MSG_LEN 1024
4576 	/* mtu * (2 ^ lp_pktn_ini) should be in the range of 1024 to mtu */
4577 	lp_pktn_ini = ilog2(max(mtu, MIN_LP_MSG_LEN) / mtu);
4578 
4579 	if (attr_mask & IB_QP_PATH_MTU) {
4580 		hr_reg_write(context, QPC_MTU, ib_mtu);
4581 		hr_reg_clear(qpc_mask, QPC_MTU);
4582 	}
4583 
4584 	hr_reg_write(context, QPC_LP_PKTN_INI, lp_pktn_ini);
4585 	hr_reg_clear(qpc_mask, QPC_LP_PKTN_INI);
4586 
4587 	/* ACK_REQ_FREQ should be larger than or equal to LP_PKTN_INI */
4588 	hr_reg_write(context, QPC_ACK_REQ_FREQ, lp_pktn_ini);
4589 	hr_reg_clear(qpc_mask, QPC_ACK_REQ_FREQ);
4590 
4591 	hr_reg_clear(qpc_mask, QPC_RX_REQ_PSN_ERR);
4592 	hr_reg_clear(qpc_mask, QPC_RX_REQ_MSN);
4593 	hr_reg_clear(qpc_mask, QPC_RX_REQ_LAST_OPTYPE);
4594 
4595 	context->rq_rnr_timer = 0;
4596 	qpc_mask->rq_rnr_timer = 0;
4597 
4598 	hr_reg_clear(qpc_mask, QPC_TRRL_HEAD_MAX);
4599 	hr_reg_clear(qpc_mask, QPC_TRRL_TAIL_MAX);
4600 
4601 	/* rocee send 2^lp_sgen_ini segs every time */
4602 	hr_reg_write(context, QPC_LP_SGEN_INI, 3);
4603 	hr_reg_clear(qpc_mask, QPC_LP_SGEN_INI);
4604 
4605 	if (udata && ibqp->qp_type == IB_QPT_RC &&
4606 	    (uctx->config & HNS_ROCE_RQ_INLINE_FLAGS)) {
4607 		hr_reg_write_bool(context, QPC_RQIE,
4608 				  hr_dev->caps.flags &
4609 				  HNS_ROCE_CAP_FLAG_RQ_INLINE);
4610 		hr_reg_clear(qpc_mask, QPC_RQIE);
4611 	}
4612 
4613 	if (udata &&
4614 	    (ibqp->qp_type == IB_QPT_RC || ibqp->qp_type == IB_QPT_XRC_TGT) &&
4615 	    (uctx->config & HNS_ROCE_CQE_INLINE_FLAGS)) {
4616 		hr_reg_write_bool(context, QPC_CQEIE,
4617 				  hr_dev->caps.flags &
4618 				  HNS_ROCE_CAP_FLAG_CQE_INLINE);
4619 		hr_reg_clear(qpc_mask, QPC_CQEIE);
4620 
4621 		hr_reg_write(context, QPC_CQEIS, 0);
4622 		hr_reg_clear(qpc_mask, QPC_CQEIS);
4623 	}
4624 
4625 	return 0;
4626 }
4627 
4628 static int modify_qp_rtr_to_rts(struct ib_qp *ibqp,
4629 				const struct ib_qp_attr *attr, int attr_mask,
4630 				struct hns_roce_v2_qp_context *context,
4631 				struct hns_roce_v2_qp_context *qpc_mask)
4632 {
4633 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4634 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4635 	struct ib_device *ibdev = &hr_dev->ib_dev;
4636 	int ret;
4637 
4638 	/* Not support alternate path and path migration */
4639 	if (attr_mask & (IB_QP_ALT_PATH | IB_QP_PATH_MIG_STATE)) {
4640 		ibdev_err(ibdev, "RTR2RTS attr_mask (0x%x)error\n", attr_mask);
4641 		return -EINVAL;
4642 	}
4643 
4644 	ret = config_qp_sq_buf(hr_dev, hr_qp, context, qpc_mask);
4645 	if (ret) {
4646 		ibdev_err(ibdev, "failed to config sq buf, ret = %d.\n", ret);
4647 		return ret;
4648 	}
4649 
4650 	/*
4651 	 * Set some fields in context to zero, Because the default values
4652 	 * of all fields in context are zero, we need not set them to 0 again.
4653 	 * but we should set the relevant fields of context mask to 0.
4654 	 */
4655 	hr_reg_clear(qpc_mask, QPC_IRRL_SGE_IDX);
4656 
4657 	hr_reg_clear(qpc_mask, QPC_RX_ACK_MSN);
4658 
4659 	hr_reg_clear(qpc_mask, QPC_ACK_LAST_OPTYPE);
4660 	hr_reg_clear(qpc_mask, QPC_IRRL_PSN_VLD);
4661 	hr_reg_clear(qpc_mask, QPC_IRRL_PSN);
4662 
4663 	hr_reg_clear(qpc_mask, QPC_IRRL_TAIL_REAL);
4664 
4665 	hr_reg_clear(qpc_mask, QPC_RETRY_MSG_MSN);
4666 
4667 	hr_reg_clear(qpc_mask, QPC_RNR_RETRY_FLAG);
4668 
4669 	hr_reg_clear(qpc_mask, QPC_CHECK_FLG);
4670 
4671 	hr_reg_clear(qpc_mask, QPC_V2_IRRL_HEAD);
4672 
4673 	return 0;
4674 }
4675 
4676 static int get_dip_ctx_idx(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
4677 			   u32 *dip_idx)
4678 {
4679 	const struct ib_global_route *grh = rdma_ah_read_grh(&attr->ah_attr);
4680 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4681 	u32 *spare_idx = hr_dev->qp_table.idx_table.spare_idx;
4682 	u32 *head =  &hr_dev->qp_table.idx_table.head;
4683 	u32 *tail =  &hr_dev->qp_table.idx_table.tail;
4684 	struct hns_roce_dip *hr_dip;
4685 	unsigned long flags;
4686 	int ret = 0;
4687 
4688 	spin_lock_irqsave(&hr_dev->dip_list_lock, flags);
4689 
4690 	spare_idx[*tail] = ibqp->qp_num;
4691 	*tail = (*tail == hr_dev->caps.num_qps - 1) ? 0 : (*tail + 1);
4692 
4693 	list_for_each_entry(hr_dip, &hr_dev->dip_list, node) {
4694 		if (!memcmp(grh->dgid.raw, hr_dip->dgid, 16)) {
4695 			*dip_idx = hr_dip->dip_idx;
4696 			goto out;
4697 		}
4698 	}
4699 
4700 	/* If no dgid is found, a new dip and a mapping between dgid and
4701 	 * dip_idx will be created.
4702 	 */
4703 	hr_dip = kzalloc(sizeof(*hr_dip), GFP_ATOMIC);
4704 	if (!hr_dip) {
4705 		ret = -ENOMEM;
4706 		goto out;
4707 	}
4708 
4709 	memcpy(hr_dip->dgid, grh->dgid.raw, sizeof(grh->dgid.raw));
4710 	hr_dip->dip_idx = *dip_idx = spare_idx[*head];
4711 	*head = (*head == hr_dev->caps.num_qps - 1) ? 0 : (*head + 1);
4712 	list_add_tail(&hr_dip->node, &hr_dev->dip_list);
4713 
4714 out:
4715 	spin_unlock_irqrestore(&hr_dev->dip_list_lock, flags);
4716 	return ret;
4717 }
4718 
4719 enum {
4720 	CONG_DCQCN,
4721 	CONG_WINDOW,
4722 };
4723 
4724 enum {
4725 	UNSUPPORT_CONG_LEVEL,
4726 	SUPPORT_CONG_LEVEL,
4727 };
4728 
4729 enum {
4730 	CONG_LDCP,
4731 	CONG_HC3,
4732 };
4733 
4734 enum {
4735 	DIP_INVALID,
4736 	DIP_VALID,
4737 };
4738 
4739 enum {
4740 	WND_LIMIT,
4741 	WND_UNLIMIT,
4742 };
4743 
4744 static int check_cong_type(struct ib_qp *ibqp,
4745 			   struct hns_roce_congestion_algorithm *cong_alg)
4746 {
4747 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4748 
4749 	if (ibqp->qp_type == IB_QPT_UD)
4750 		hr_dev->caps.cong_type = CONG_TYPE_DCQCN;
4751 
4752 	/* different congestion types match different configurations */
4753 	switch (hr_dev->caps.cong_type) {
4754 	case CONG_TYPE_DCQCN:
4755 		cong_alg->alg_sel = CONG_DCQCN;
4756 		cong_alg->alg_sub_sel = UNSUPPORT_CONG_LEVEL;
4757 		cong_alg->dip_vld = DIP_INVALID;
4758 		cong_alg->wnd_mode_sel = WND_LIMIT;
4759 		break;
4760 	case CONG_TYPE_LDCP:
4761 		cong_alg->alg_sel = CONG_WINDOW;
4762 		cong_alg->alg_sub_sel = CONG_LDCP;
4763 		cong_alg->dip_vld = DIP_INVALID;
4764 		cong_alg->wnd_mode_sel = WND_UNLIMIT;
4765 		break;
4766 	case CONG_TYPE_HC3:
4767 		cong_alg->alg_sel = CONG_WINDOW;
4768 		cong_alg->alg_sub_sel = CONG_HC3;
4769 		cong_alg->dip_vld = DIP_INVALID;
4770 		cong_alg->wnd_mode_sel = WND_LIMIT;
4771 		break;
4772 	case CONG_TYPE_DIP:
4773 		cong_alg->alg_sel = CONG_DCQCN;
4774 		cong_alg->alg_sub_sel = UNSUPPORT_CONG_LEVEL;
4775 		cong_alg->dip_vld = DIP_VALID;
4776 		cong_alg->wnd_mode_sel = WND_LIMIT;
4777 		break;
4778 	default:
4779 		ibdev_warn(&hr_dev->ib_dev,
4780 			   "invalid type(%u) for congestion selection.\n",
4781 			   hr_dev->caps.cong_type);
4782 		hr_dev->caps.cong_type = CONG_TYPE_DCQCN;
4783 		cong_alg->alg_sel = CONG_DCQCN;
4784 		cong_alg->alg_sub_sel = UNSUPPORT_CONG_LEVEL;
4785 		cong_alg->dip_vld = DIP_INVALID;
4786 		cong_alg->wnd_mode_sel = WND_LIMIT;
4787 		break;
4788 	}
4789 
4790 	return 0;
4791 }
4792 
4793 static int fill_cong_field(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
4794 			   struct hns_roce_v2_qp_context *context,
4795 			   struct hns_roce_v2_qp_context *qpc_mask)
4796 {
4797 	const struct ib_global_route *grh = rdma_ah_read_grh(&attr->ah_attr);
4798 	struct hns_roce_congestion_algorithm cong_field;
4799 	struct ib_device *ibdev = ibqp->device;
4800 	struct hns_roce_dev *hr_dev = to_hr_dev(ibdev);
4801 	u32 dip_idx = 0;
4802 	int ret;
4803 
4804 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08 ||
4805 	    grh->sgid_attr->gid_type == IB_GID_TYPE_ROCE)
4806 		return 0;
4807 
4808 	ret = check_cong_type(ibqp, &cong_field);
4809 	if (ret)
4810 		return ret;
4811 
4812 	hr_reg_write(context, QPC_CONG_ALGO_TMPL_ID, hr_dev->cong_algo_tmpl_id +
4813 		     hr_dev->caps.cong_type * HNS_ROCE_CONG_SIZE);
4814 	hr_reg_clear(qpc_mask, QPC_CONG_ALGO_TMPL_ID);
4815 	hr_reg_write(&context->ext, QPCEX_CONG_ALG_SEL, cong_field.alg_sel);
4816 	hr_reg_clear(&qpc_mask->ext, QPCEX_CONG_ALG_SEL);
4817 	hr_reg_write(&context->ext, QPCEX_CONG_ALG_SUB_SEL,
4818 		     cong_field.alg_sub_sel);
4819 	hr_reg_clear(&qpc_mask->ext, QPCEX_CONG_ALG_SUB_SEL);
4820 	hr_reg_write(&context->ext, QPCEX_DIP_CTX_IDX_VLD, cong_field.dip_vld);
4821 	hr_reg_clear(&qpc_mask->ext, QPCEX_DIP_CTX_IDX_VLD);
4822 	hr_reg_write(&context->ext, QPCEX_SQ_RQ_NOT_FORBID_EN,
4823 		     cong_field.wnd_mode_sel);
4824 	hr_reg_clear(&qpc_mask->ext, QPCEX_SQ_RQ_NOT_FORBID_EN);
4825 
4826 	/* if dip is disabled, there is no need to set dip idx */
4827 	if (cong_field.dip_vld == 0)
4828 		return 0;
4829 
4830 	ret = get_dip_ctx_idx(ibqp, attr, &dip_idx);
4831 	if (ret) {
4832 		ibdev_err(ibdev, "failed to fill cong field, ret = %d.\n", ret);
4833 		return ret;
4834 	}
4835 
4836 	hr_reg_write(&context->ext, QPCEX_DIP_CTX_IDX, dip_idx);
4837 	hr_reg_write(&qpc_mask->ext, QPCEX_DIP_CTX_IDX, 0);
4838 
4839 	return 0;
4840 }
4841 
4842 static int hns_roce_v2_set_path(struct ib_qp *ibqp,
4843 				const struct ib_qp_attr *attr,
4844 				int attr_mask,
4845 				struct hns_roce_v2_qp_context *context,
4846 				struct hns_roce_v2_qp_context *qpc_mask)
4847 {
4848 	const struct ib_global_route *grh = rdma_ah_read_grh(&attr->ah_attr);
4849 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4850 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4851 	struct ib_device *ibdev = &hr_dev->ib_dev;
4852 	const struct ib_gid_attr *gid_attr = NULL;
4853 	u8 sl = rdma_ah_get_sl(&attr->ah_attr);
4854 	int is_roce_protocol;
4855 	u16 vlan_id = 0xffff;
4856 	bool is_udp = false;
4857 	u32 max_sl;
4858 	u8 ib_port;
4859 	u8 hr_port;
4860 	int ret;
4861 
4862 	max_sl = min_t(u32, MAX_SERVICE_LEVEL, hr_dev->caps.sl_num - 1);
4863 	if (unlikely(sl > max_sl)) {
4864 		ibdev_err_ratelimited(ibdev,
4865 				      "failed to fill QPC, sl (%u) shouldn't be larger than %u.\n",
4866 				      sl, max_sl);
4867 		return -EINVAL;
4868 	}
4869 
4870 	/*
4871 	 * If free_mr_en of qp is set, it means that this qp comes from
4872 	 * free mr. This qp will perform the loopback operation.
4873 	 * In the loopback scenario, only sl needs to be set.
4874 	 */
4875 	if (hr_qp->free_mr_en) {
4876 		hr_reg_write(context, QPC_SL, sl);
4877 		hr_reg_clear(qpc_mask, QPC_SL);
4878 		hr_qp->sl = sl;
4879 		return 0;
4880 	}
4881 
4882 	ib_port = (attr_mask & IB_QP_PORT) ? attr->port_num : hr_qp->port + 1;
4883 	hr_port = ib_port - 1;
4884 	is_roce_protocol = rdma_cap_eth_ah(&hr_dev->ib_dev, ib_port) &&
4885 			   rdma_ah_get_ah_flags(&attr->ah_attr) & IB_AH_GRH;
4886 
4887 	if (is_roce_protocol) {
4888 		gid_attr = attr->ah_attr.grh.sgid_attr;
4889 		ret = rdma_read_gid_l2_fields(gid_attr, &vlan_id, NULL);
4890 		if (ret)
4891 			return ret;
4892 
4893 		is_udp = (gid_attr->gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP);
4894 	}
4895 
4896 	/* Only HIP08 needs to set the vlan_en bits in QPC */
4897 	if (vlan_id < VLAN_N_VID &&
4898 	    hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
4899 		hr_reg_enable(context, QPC_RQ_VLAN_EN);
4900 		hr_reg_clear(qpc_mask, QPC_RQ_VLAN_EN);
4901 		hr_reg_enable(context, QPC_SQ_VLAN_EN);
4902 		hr_reg_clear(qpc_mask, QPC_SQ_VLAN_EN);
4903 	}
4904 
4905 	hr_reg_write(context, QPC_VLAN_ID, vlan_id);
4906 	hr_reg_clear(qpc_mask, QPC_VLAN_ID);
4907 
4908 	if (grh->sgid_index >= hr_dev->caps.gid_table_len[hr_port]) {
4909 		ibdev_err(ibdev, "sgid_index(%u) too large. max is %d\n",
4910 			  grh->sgid_index, hr_dev->caps.gid_table_len[hr_port]);
4911 		return -EINVAL;
4912 	}
4913 
4914 	if (attr->ah_attr.type != RDMA_AH_ATTR_TYPE_ROCE) {
4915 		ibdev_err(ibdev, "ah attr is not RDMA roce type\n");
4916 		return -EINVAL;
4917 	}
4918 
4919 	hr_reg_write(context, QPC_UDPSPN,
4920 		     is_udp ? rdma_get_udp_sport(grh->flow_label, ibqp->qp_num,
4921 						 attr->dest_qp_num) :
4922 				    0);
4923 
4924 	hr_reg_clear(qpc_mask, QPC_UDPSPN);
4925 
4926 	hr_reg_write(context, QPC_GMV_IDX, grh->sgid_index);
4927 
4928 	hr_reg_clear(qpc_mask, QPC_GMV_IDX);
4929 
4930 	hr_reg_write(context, QPC_HOPLIMIT, grh->hop_limit);
4931 	hr_reg_clear(qpc_mask, QPC_HOPLIMIT);
4932 
4933 	ret = fill_cong_field(ibqp, attr, context, qpc_mask);
4934 	if (ret)
4935 		return ret;
4936 
4937 	hr_reg_write(context, QPC_TC, get_tclass(&attr->ah_attr.grh));
4938 	hr_reg_clear(qpc_mask, QPC_TC);
4939 
4940 	hr_reg_write(context, QPC_FL, grh->flow_label);
4941 	hr_reg_clear(qpc_mask, QPC_FL);
4942 	memcpy(context->dgid, grh->dgid.raw, sizeof(grh->dgid.raw));
4943 	memset(qpc_mask->dgid, 0, sizeof(grh->dgid.raw));
4944 
4945 	hr_qp->sl = sl;
4946 	hr_reg_write(context, QPC_SL, hr_qp->sl);
4947 	hr_reg_clear(qpc_mask, QPC_SL);
4948 
4949 	return 0;
4950 }
4951 
4952 static bool check_qp_state(enum ib_qp_state cur_state,
4953 			   enum ib_qp_state new_state)
4954 {
4955 	static const bool sm[][IB_QPS_ERR + 1] = {
4956 		[IB_QPS_RESET] = { [IB_QPS_RESET] = true,
4957 				   [IB_QPS_INIT] = true },
4958 		[IB_QPS_INIT] = { [IB_QPS_RESET] = true,
4959 				  [IB_QPS_INIT] = true,
4960 				  [IB_QPS_RTR] = true,
4961 				  [IB_QPS_ERR] = true },
4962 		[IB_QPS_RTR] = { [IB_QPS_RESET] = true,
4963 				 [IB_QPS_RTS] = true,
4964 				 [IB_QPS_ERR] = true },
4965 		[IB_QPS_RTS] = { [IB_QPS_RESET] = true,
4966 				 [IB_QPS_RTS] = true,
4967 				 [IB_QPS_ERR] = true },
4968 		[IB_QPS_SQD] = {},
4969 		[IB_QPS_SQE] = {},
4970 		[IB_QPS_ERR] = { [IB_QPS_RESET] = true,
4971 				 [IB_QPS_ERR] = true }
4972 	};
4973 
4974 	return sm[cur_state][new_state];
4975 }
4976 
4977 static int hns_roce_v2_set_abs_fields(struct ib_qp *ibqp,
4978 				      const struct ib_qp_attr *attr,
4979 				      int attr_mask,
4980 				      enum ib_qp_state cur_state,
4981 				      enum ib_qp_state new_state,
4982 				      struct hns_roce_v2_qp_context *context,
4983 				      struct hns_roce_v2_qp_context *qpc_mask,
4984 				      struct ib_udata *udata)
4985 {
4986 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4987 	int ret = 0;
4988 
4989 	if (!check_qp_state(cur_state, new_state)) {
4990 		ibdev_err(&hr_dev->ib_dev, "Illegal state for QP!\n");
4991 		return -EINVAL;
4992 	}
4993 
4994 	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) {
4995 		memset(qpc_mask, 0, hr_dev->caps.qpc_sz);
4996 		modify_qp_reset_to_init(ibqp, attr, context, qpc_mask);
4997 	} else if (cur_state == IB_QPS_INIT && new_state == IB_QPS_INIT) {
4998 		modify_qp_init_to_init(ibqp, attr, context, qpc_mask);
4999 	} else if (cur_state == IB_QPS_INIT && new_state == IB_QPS_RTR) {
5000 		ret = modify_qp_init_to_rtr(ibqp, attr, attr_mask, context,
5001 					    qpc_mask, udata);
5002 	} else if (cur_state == IB_QPS_RTR && new_state == IB_QPS_RTS) {
5003 		ret = modify_qp_rtr_to_rts(ibqp, attr, attr_mask, context,
5004 					   qpc_mask);
5005 	}
5006 
5007 	return ret;
5008 }
5009 
5010 static bool check_qp_timeout_cfg_range(struct hns_roce_dev *hr_dev, u8 *timeout)
5011 {
5012 #define QP_ACK_TIMEOUT_MAX_HIP08 20
5013 #define QP_ACK_TIMEOUT_MAX 31
5014 
5015 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
5016 		if (*timeout > QP_ACK_TIMEOUT_MAX_HIP08) {
5017 			ibdev_warn(&hr_dev->ib_dev,
5018 				   "local ACK timeout shall be 0 to 20.\n");
5019 			return false;
5020 		}
5021 		*timeout += HNS_ROCE_V2_QP_ACK_TIMEOUT_OFS_HIP08;
5022 	} else if (hr_dev->pci_dev->revision > PCI_REVISION_ID_HIP08) {
5023 		if (*timeout > QP_ACK_TIMEOUT_MAX) {
5024 			ibdev_warn(&hr_dev->ib_dev,
5025 				   "local ACK timeout shall be 0 to 31.\n");
5026 			return false;
5027 		}
5028 	}
5029 
5030 	return true;
5031 }
5032 
5033 static int hns_roce_v2_set_opt_fields(struct ib_qp *ibqp,
5034 				      const struct ib_qp_attr *attr,
5035 				      int attr_mask,
5036 				      struct hns_roce_v2_qp_context *context,
5037 				      struct hns_roce_v2_qp_context *qpc_mask)
5038 {
5039 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5040 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5041 	int ret = 0;
5042 	u8 timeout;
5043 
5044 	if (attr_mask & IB_QP_AV) {
5045 		ret = hns_roce_v2_set_path(ibqp, attr, attr_mask, context,
5046 					   qpc_mask);
5047 		if (ret)
5048 			return ret;
5049 	}
5050 
5051 	if (attr_mask & IB_QP_TIMEOUT) {
5052 		timeout = attr->timeout;
5053 		if (check_qp_timeout_cfg_range(hr_dev, &timeout)) {
5054 			hr_reg_write(context, QPC_AT, timeout);
5055 			hr_reg_clear(qpc_mask, QPC_AT);
5056 		}
5057 	}
5058 
5059 	if (attr_mask & IB_QP_RETRY_CNT) {
5060 		hr_reg_write(context, QPC_RETRY_NUM_INIT, attr->retry_cnt);
5061 		hr_reg_clear(qpc_mask, QPC_RETRY_NUM_INIT);
5062 
5063 		hr_reg_write(context, QPC_RETRY_CNT, attr->retry_cnt);
5064 		hr_reg_clear(qpc_mask, QPC_RETRY_CNT);
5065 	}
5066 
5067 	if (attr_mask & IB_QP_RNR_RETRY) {
5068 		hr_reg_write(context, QPC_RNR_NUM_INIT, attr->rnr_retry);
5069 		hr_reg_clear(qpc_mask, QPC_RNR_NUM_INIT);
5070 
5071 		hr_reg_write(context, QPC_RNR_CNT, attr->rnr_retry);
5072 		hr_reg_clear(qpc_mask, QPC_RNR_CNT);
5073 	}
5074 
5075 	if (attr_mask & IB_QP_SQ_PSN) {
5076 		hr_reg_write(context, QPC_SQ_CUR_PSN, attr->sq_psn);
5077 		hr_reg_clear(qpc_mask, QPC_SQ_CUR_PSN);
5078 
5079 		hr_reg_write(context, QPC_SQ_MAX_PSN, attr->sq_psn);
5080 		hr_reg_clear(qpc_mask, QPC_SQ_MAX_PSN);
5081 
5082 		hr_reg_write(context, QPC_RETRY_MSG_PSN_L, attr->sq_psn);
5083 		hr_reg_clear(qpc_mask, QPC_RETRY_MSG_PSN_L);
5084 
5085 		hr_reg_write(context, QPC_RETRY_MSG_PSN_H,
5086 			     attr->sq_psn >> RETRY_MSG_PSN_SHIFT);
5087 		hr_reg_clear(qpc_mask, QPC_RETRY_MSG_PSN_H);
5088 
5089 		hr_reg_write(context, QPC_RETRY_MSG_FPKT_PSN, attr->sq_psn);
5090 		hr_reg_clear(qpc_mask, QPC_RETRY_MSG_FPKT_PSN);
5091 
5092 		hr_reg_write(context, QPC_RX_ACK_EPSN, attr->sq_psn);
5093 		hr_reg_clear(qpc_mask, QPC_RX_ACK_EPSN);
5094 	}
5095 
5096 	if ((attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) &&
5097 	     attr->max_dest_rd_atomic) {
5098 		hr_reg_write(context, QPC_RR_MAX,
5099 			     fls(attr->max_dest_rd_atomic - 1));
5100 		hr_reg_clear(qpc_mask, QPC_RR_MAX);
5101 	}
5102 
5103 	if ((attr_mask & IB_QP_MAX_QP_RD_ATOMIC) && attr->max_rd_atomic) {
5104 		hr_reg_write(context, QPC_SR_MAX, fls(attr->max_rd_atomic - 1));
5105 		hr_reg_clear(qpc_mask, QPC_SR_MAX);
5106 	}
5107 
5108 	if (attr_mask & (IB_QP_ACCESS_FLAGS | IB_QP_MAX_DEST_RD_ATOMIC))
5109 		set_access_flags(hr_qp, context, qpc_mask, attr, attr_mask);
5110 
5111 	if (attr_mask & IB_QP_MIN_RNR_TIMER) {
5112 		hr_reg_write(context, QPC_MIN_RNR_TIME,
5113 			    hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08 ?
5114 			    HNS_ROCE_RNR_TIMER_10NS : attr->min_rnr_timer);
5115 		hr_reg_clear(qpc_mask, QPC_MIN_RNR_TIME);
5116 	}
5117 
5118 	if (attr_mask & IB_QP_RQ_PSN) {
5119 		hr_reg_write(context, QPC_RX_REQ_EPSN, attr->rq_psn);
5120 		hr_reg_clear(qpc_mask, QPC_RX_REQ_EPSN);
5121 
5122 		hr_reg_write(context, QPC_RAQ_PSN, attr->rq_psn - 1);
5123 		hr_reg_clear(qpc_mask, QPC_RAQ_PSN);
5124 	}
5125 
5126 	if (attr_mask & IB_QP_QKEY) {
5127 		context->qkey_xrcd = cpu_to_le32(attr->qkey);
5128 		qpc_mask->qkey_xrcd = 0;
5129 		hr_qp->qkey = attr->qkey;
5130 	}
5131 
5132 	return ret;
5133 }
5134 
5135 static void hns_roce_v2_record_opt_fields(struct ib_qp *ibqp,
5136 					  const struct ib_qp_attr *attr,
5137 					  int attr_mask)
5138 {
5139 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5140 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5141 
5142 	if (attr_mask & IB_QP_ACCESS_FLAGS)
5143 		hr_qp->atomic_rd_en = attr->qp_access_flags;
5144 
5145 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
5146 		hr_qp->resp_depth = attr->max_dest_rd_atomic;
5147 	if (attr_mask & IB_QP_PORT) {
5148 		hr_qp->port = attr->port_num - 1;
5149 		hr_qp->phy_port = hr_dev->iboe.phy_port[hr_qp->port];
5150 	}
5151 }
5152 
5153 static void clear_qp(struct hns_roce_qp *hr_qp)
5154 {
5155 	struct ib_qp *ibqp = &hr_qp->ibqp;
5156 
5157 	if (ibqp->send_cq)
5158 		hns_roce_v2_cq_clean(to_hr_cq(ibqp->send_cq),
5159 				     hr_qp->qpn, NULL);
5160 
5161 	if (ibqp->recv_cq  && ibqp->recv_cq != ibqp->send_cq)
5162 		hns_roce_v2_cq_clean(to_hr_cq(ibqp->recv_cq),
5163 				     hr_qp->qpn, ibqp->srq ?
5164 				     to_hr_srq(ibqp->srq) : NULL);
5165 
5166 	if (hr_qp->en_flags & HNS_ROCE_QP_CAP_RQ_RECORD_DB)
5167 		*hr_qp->rdb.db_record = 0;
5168 
5169 	hr_qp->rq.head = 0;
5170 	hr_qp->rq.tail = 0;
5171 	hr_qp->sq.head = 0;
5172 	hr_qp->sq.tail = 0;
5173 	hr_qp->next_sge = 0;
5174 }
5175 
5176 static void v2_set_flushed_fields(struct ib_qp *ibqp,
5177 				  struct hns_roce_v2_qp_context *context,
5178 				  struct hns_roce_v2_qp_context *qpc_mask)
5179 {
5180 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5181 	unsigned long sq_flag = 0;
5182 	unsigned long rq_flag = 0;
5183 
5184 	if (ibqp->qp_type == IB_QPT_XRC_TGT)
5185 		return;
5186 
5187 	spin_lock_irqsave(&hr_qp->sq.lock, sq_flag);
5188 	hr_reg_write(context, QPC_SQ_PRODUCER_IDX, hr_qp->sq.head);
5189 	hr_reg_clear(qpc_mask, QPC_SQ_PRODUCER_IDX);
5190 	hr_qp->state = IB_QPS_ERR;
5191 	spin_unlock_irqrestore(&hr_qp->sq.lock, sq_flag);
5192 
5193 	if (ibqp->srq || ibqp->qp_type == IB_QPT_XRC_INI) /* no RQ */
5194 		return;
5195 
5196 	spin_lock_irqsave(&hr_qp->rq.lock, rq_flag);
5197 	hr_reg_write(context, QPC_RQ_PRODUCER_IDX, hr_qp->rq.head);
5198 	hr_reg_clear(qpc_mask, QPC_RQ_PRODUCER_IDX);
5199 	spin_unlock_irqrestore(&hr_qp->rq.lock, rq_flag);
5200 }
5201 
5202 static int hns_roce_v2_modify_qp(struct ib_qp *ibqp,
5203 				 const struct ib_qp_attr *attr,
5204 				 int attr_mask, enum ib_qp_state cur_state,
5205 				 enum ib_qp_state new_state, struct ib_udata *udata)
5206 {
5207 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5208 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5209 	struct hns_roce_v2_qp_context ctx[2];
5210 	struct hns_roce_v2_qp_context *context = ctx;
5211 	struct hns_roce_v2_qp_context *qpc_mask = ctx + 1;
5212 	struct ib_device *ibdev = &hr_dev->ib_dev;
5213 	int ret;
5214 
5215 	if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
5216 		return -EOPNOTSUPP;
5217 
5218 	/*
5219 	 * In v2 engine, software pass context and context mask to hardware
5220 	 * when modifying qp. If software need modify some fields in context,
5221 	 * we should set all bits of the relevant fields in context mask to
5222 	 * 0 at the same time, else set them to 0x1.
5223 	 */
5224 	memset(context, 0, hr_dev->caps.qpc_sz);
5225 	memset(qpc_mask, 0xff, hr_dev->caps.qpc_sz);
5226 
5227 	ret = hns_roce_v2_set_abs_fields(ibqp, attr, attr_mask, cur_state,
5228 					 new_state, context, qpc_mask, udata);
5229 	if (ret)
5230 		goto out;
5231 
5232 	/* When QP state is err, SQ and RQ WQE should be flushed */
5233 	if (new_state == IB_QPS_ERR)
5234 		v2_set_flushed_fields(ibqp, context, qpc_mask);
5235 
5236 	/* Configure the optional fields */
5237 	ret = hns_roce_v2_set_opt_fields(ibqp, attr, attr_mask, context,
5238 					 qpc_mask);
5239 	if (ret)
5240 		goto out;
5241 
5242 	hr_reg_write_bool(context, QPC_INV_CREDIT,
5243 			  to_hr_qp_type(hr_qp->ibqp.qp_type) == SERV_TYPE_XRC ||
5244 			  ibqp->srq);
5245 	hr_reg_clear(qpc_mask, QPC_INV_CREDIT);
5246 
5247 	/* Every status migrate must change state */
5248 	hr_reg_write(context, QPC_QP_ST, new_state);
5249 	hr_reg_clear(qpc_mask, QPC_QP_ST);
5250 
5251 	/* SW pass context to HW */
5252 	ret = hns_roce_v2_qp_modify(hr_dev, context, qpc_mask, hr_qp);
5253 	if (ret) {
5254 		ibdev_err(ibdev, "failed to modify QP, ret = %d.\n", ret);
5255 		goto out;
5256 	}
5257 
5258 	hr_qp->state = new_state;
5259 
5260 	hns_roce_v2_record_opt_fields(ibqp, attr, attr_mask);
5261 
5262 	if (new_state == IB_QPS_RESET && !ibqp->uobject)
5263 		clear_qp(hr_qp);
5264 
5265 out:
5266 	return ret;
5267 }
5268 
5269 static int to_ib_qp_st(enum hns_roce_v2_qp_state state)
5270 {
5271 	static const enum ib_qp_state map[] = {
5272 		[HNS_ROCE_QP_ST_RST] = IB_QPS_RESET,
5273 		[HNS_ROCE_QP_ST_INIT] = IB_QPS_INIT,
5274 		[HNS_ROCE_QP_ST_RTR] = IB_QPS_RTR,
5275 		[HNS_ROCE_QP_ST_RTS] = IB_QPS_RTS,
5276 		[HNS_ROCE_QP_ST_SQD] = IB_QPS_SQD,
5277 		[HNS_ROCE_QP_ST_SQER] = IB_QPS_SQE,
5278 		[HNS_ROCE_QP_ST_ERR] = IB_QPS_ERR,
5279 		[HNS_ROCE_QP_ST_SQ_DRAINING] = IB_QPS_SQD
5280 	};
5281 
5282 	return (state < ARRAY_SIZE(map)) ? map[state] : -1;
5283 }
5284 
5285 static int hns_roce_v2_query_qpc(struct hns_roce_dev *hr_dev, u32 qpn,
5286 				 void *buffer)
5287 {
5288 	struct hns_roce_cmd_mailbox *mailbox;
5289 	int ret;
5290 
5291 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5292 	if (IS_ERR(mailbox))
5293 		return PTR_ERR(mailbox);
5294 
5295 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, HNS_ROCE_CMD_QUERY_QPC,
5296 				qpn);
5297 	if (ret)
5298 		goto out;
5299 
5300 	memcpy(buffer, mailbox->buf, hr_dev->caps.qpc_sz);
5301 
5302 out:
5303 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5304 	return ret;
5305 }
5306 
5307 static int hns_roce_v2_query_srqc(struct hns_roce_dev *hr_dev, u32 srqn,
5308 				 void *buffer)
5309 {
5310 	struct hns_roce_srq_context *context;
5311 	struct hns_roce_cmd_mailbox *mailbox;
5312 	int ret;
5313 
5314 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5315 	if (IS_ERR(mailbox))
5316 		return PTR_ERR(mailbox);
5317 
5318 	context = mailbox->buf;
5319 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, HNS_ROCE_CMD_QUERY_SRQC,
5320 				srqn);
5321 	if (ret)
5322 		goto out;
5323 
5324 	memcpy(buffer, context, sizeof(*context));
5325 
5326 out:
5327 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5328 	return ret;
5329 }
5330 
5331 static u8 get_qp_timeout_attr(struct hns_roce_dev *hr_dev,
5332 			      struct hns_roce_v2_qp_context *context)
5333 {
5334 	u8 timeout;
5335 
5336 	timeout = (u8)hr_reg_read(context, QPC_AT);
5337 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08)
5338 		timeout -= HNS_ROCE_V2_QP_ACK_TIMEOUT_OFS_HIP08;
5339 
5340 	return timeout;
5341 }
5342 
5343 static int hns_roce_v2_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
5344 				int qp_attr_mask,
5345 				struct ib_qp_init_attr *qp_init_attr)
5346 {
5347 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5348 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5349 	struct hns_roce_v2_qp_context context = {};
5350 	struct ib_device *ibdev = &hr_dev->ib_dev;
5351 	int tmp_qp_state;
5352 	int state;
5353 	int ret;
5354 
5355 	memset(qp_attr, 0, sizeof(*qp_attr));
5356 	memset(qp_init_attr, 0, sizeof(*qp_init_attr));
5357 
5358 	mutex_lock(&hr_qp->mutex);
5359 
5360 	if (hr_qp->state == IB_QPS_RESET) {
5361 		qp_attr->qp_state = IB_QPS_RESET;
5362 		ret = 0;
5363 		goto done;
5364 	}
5365 
5366 	ret = hns_roce_v2_query_qpc(hr_dev, hr_qp->qpn, &context);
5367 	if (ret) {
5368 		ibdev_err(ibdev, "failed to query QPC, ret = %d.\n", ret);
5369 		ret = -EINVAL;
5370 		goto out;
5371 	}
5372 
5373 	state = hr_reg_read(&context, QPC_QP_ST);
5374 	tmp_qp_state = to_ib_qp_st((enum hns_roce_v2_qp_state)state);
5375 	if (tmp_qp_state == -1) {
5376 		ibdev_err(ibdev, "Illegal ib_qp_state\n");
5377 		ret = -EINVAL;
5378 		goto out;
5379 	}
5380 	hr_qp->state = (u8)tmp_qp_state;
5381 	qp_attr->qp_state = (enum ib_qp_state)hr_qp->state;
5382 	qp_attr->path_mtu = (enum ib_mtu)hr_reg_read(&context, QPC_MTU);
5383 	qp_attr->path_mig_state = IB_MIG_ARMED;
5384 	qp_attr->ah_attr.type = RDMA_AH_ATTR_TYPE_ROCE;
5385 	if (hr_qp->ibqp.qp_type == IB_QPT_UD)
5386 		qp_attr->qkey = le32_to_cpu(context.qkey_xrcd);
5387 
5388 	qp_attr->rq_psn = hr_reg_read(&context, QPC_RX_REQ_EPSN);
5389 	qp_attr->sq_psn = (u32)hr_reg_read(&context, QPC_SQ_CUR_PSN);
5390 	qp_attr->dest_qp_num = hr_reg_read(&context, QPC_DQPN);
5391 	qp_attr->qp_access_flags =
5392 		((hr_reg_read(&context, QPC_RRE)) << V2_QP_RRE_S) |
5393 		((hr_reg_read(&context, QPC_RWE)) << V2_QP_RWE_S) |
5394 		((hr_reg_read(&context, QPC_ATE)) << V2_QP_ATE_S);
5395 
5396 	if (hr_qp->ibqp.qp_type == IB_QPT_RC ||
5397 	    hr_qp->ibqp.qp_type == IB_QPT_XRC_INI ||
5398 	    hr_qp->ibqp.qp_type == IB_QPT_XRC_TGT) {
5399 		struct ib_global_route *grh =
5400 			rdma_ah_retrieve_grh(&qp_attr->ah_attr);
5401 
5402 		rdma_ah_set_sl(&qp_attr->ah_attr,
5403 			       hr_reg_read(&context, QPC_SL));
5404 		rdma_ah_set_port_num(&qp_attr->ah_attr, hr_qp->port + 1);
5405 		rdma_ah_set_ah_flags(&qp_attr->ah_attr, IB_AH_GRH);
5406 		grh->flow_label = hr_reg_read(&context, QPC_FL);
5407 		grh->sgid_index = hr_reg_read(&context, QPC_GMV_IDX);
5408 		grh->hop_limit = hr_reg_read(&context, QPC_HOPLIMIT);
5409 		grh->traffic_class = hr_reg_read(&context, QPC_TC);
5410 
5411 		memcpy(grh->dgid.raw, context.dgid, sizeof(grh->dgid.raw));
5412 	}
5413 
5414 	qp_attr->port_num = hr_qp->port + 1;
5415 	qp_attr->sq_draining = 0;
5416 	qp_attr->max_rd_atomic = 1 << hr_reg_read(&context, QPC_SR_MAX);
5417 	qp_attr->max_dest_rd_atomic = 1 << hr_reg_read(&context, QPC_RR_MAX);
5418 
5419 	qp_attr->min_rnr_timer = (u8)hr_reg_read(&context, QPC_MIN_RNR_TIME);
5420 	qp_attr->timeout = get_qp_timeout_attr(hr_dev, &context);
5421 	qp_attr->retry_cnt = hr_reg_read(&context, QPC_RETRY_NUM_INIT);
5422 	qp_attr->rnr_retry = hr_reg_read(&context, QPC_RNR_NUM_INIT);
5423 
5424 done:
5425 	qp_attr->cur_qp_state = qp_attr->qp_state;
5426 	qp_attr->cap.max_recv_wr = hr_qp->rq.wqe_cnt;
5427 	qp_attr->cap.max_recv_sge = hr_qp->rq.max_gs - hr_qp->rq.rsv_sge;
5428 	qp_attr->cap.max_inline_data = hr_qp->max_inline_data;
5429 
5430 	qp_attr->cap.max_send_wr = hr_qp->sq.wqe_cnt;
5431 	qp_attr->cap.max_send_sge = hr_qp->sq.max_gs;
5432 
5433 	qp_init_attr->qp_context = ibqp->qp_context;
5434 	qp_init_attr->qp_type = ibqp->qp_type;
5435 	qp_init_attr->recv_cq = ibqp->recv_cq;
5436 	qp_init_attr->send_cq = ibqp->send_cq;
5437 	qp_init_attr->srq = ibqp->srq;
5438 	qp_init_attr->cap = qp_attr->cap;
5439 	qp_init_attr->sq_sig_type = hr_qp->sq_signal_bits;
5440 
5441 out:
5442 	mutex_unlock(&hr_qp->mutex);
5443 	return ret;
5444 }
5445 
5446 static inline int modify_qp_is_ok(struct hns_roce_qp *hr_qp)
5447 {
5448 	return ((hr_qp->ibqp.qp_type == IB_QPT_RC ||
5449 		 hr_qp->ibqp.qp_type == IB_QPT_UD ||
5450 		 hr_qp->ibqp.qp_type == IB_QPT_XRC_INI ||
5451 		 hr_qp->ibqp.qp_type == IB_QPT_XRC_TGT) &&
5452 		hr_qp->state != IB_QPS_RESET);
5453 }
5454 
5455 static int hns_roce_v2_destroy_qp_common(struct hns_roce_dev *hr_dev,
5456 					 struct hns_roce_qp *hr_qp,
5457 					 struct ib_udata *udata)
5458 {
5459 	struct ib_device *ibdev = &hr_dev->ib_dev;
5460 	struct hns_roce_cq *send_cq, *recv_cq;
5461 	unsigned long flags;
5462 	int ret = 0;
5463 
5464 	if (modify_qp_is_ok(hr_qp)) {
5465 		/* Modify qp to reset before destroying qp */
5466 		ret = hns_roce_v2_modify_qp(&hr_qp->ibqp, NULL, 0,
5467 					    hr_qp->state, IB_QPS_RESET, udata);
5468 		if (ret)
5469 			ibdev_err(ibdev,
5470 				  "failed to modify QP to RST, ret = %d.\n",
5471 				  ret);
5472 	}
5473 
5474 	send_cq = hr_qp->ibqp.send_cq ? to_hr_cq(hr_qp->ibqp.send_cq) : NULL;
5475 	recv_cq = hr_qp->ibqp.recv_cq ? to_hr_cq(hr_qp->ibqp.recv_cq) : NULL;
5476 
5477 	spin_lock_irqsave(&hr_dev->qp_list_lock, flags);
5478 	hns_roce_lock_cqs(send_cq, recv_cq);
5479 
5480 	if (!udata) {
5481 		if (recv_cq)
5482 			__hns_roce_v2_cq_clean(recv_cq, hr_qp->qpn,
5483 					       (hr_qp->ibqp.srq ?
5484 						to_hr_srq(hr_qp->ibqp.srq) :
5485 						NULL));
5486 
5487 		if (send_cq && send_cq != recv_cq)
5488 			__hns_roce_v2_cq_clean(send_cq, hr_qp->qpn, NULL);
5489 	}
5490 
5491 	hns_roce_qp_remove(hr_dev, hr_qp);
5492 
5493 	hns_roce_unlock_cqs(send_cq, recv_cq);
5494 	spin_unlock_irqrestore(&hr_dev->qp_list_lock, flags);
5495 
5496 	return ret;
5497 }
5498 
5499 int hns_roce_v2_destroy_qp(struct ib_qp *ibqp, struct ib_udata *udata)
5500 {
5501 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5502 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5503 	int ret;
5504 
5505 	ret = hns_roce_v2_destroy_qp_common(hr_dev, hr_qp, udata);
5506 	if (ret)
5507 		ibdev_err(&hr_dev->ib_dev,
5508 			  "failed to destroy QP, QPN = 0x%06lx, ret = %d.\n",
5509 			  hr_qp->qpn, ret);
5510 
5511 	hns_roce_qp_destroy(hr_dev, hr_qp, udata);
5512 
5513 	return 0;
5514 }
5515 
5516 static int hns_roce_v2_qp_flow_control_init(struct hns_roce_dev *hr_dev,
5517 					    struct hns_roce_qp *hr_qp)
5518 {
5519 	struct ib_device *ibdev = &hr_dev->ib_dev;
5520 	struct hns_roce_sccc_clr_done *resp;
5521 	struct hns_roce_sccc_clr *clr;
5522 	struct hns_roce_cmq_desc desc;
5523 	int ret, i;
5524 
5525 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
5526 		return 0;
5527 
5528 	mutex_lock(&hr_dev->qp_table.scc_mutex);
5529 
5530 	/* set scc ctx clear done flag */
5531 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_RESET_SCCC, false);
5532 	ret =  hns_roce_cmq_send(hr_dev, &desc, 1);
5533 	if (ret) {
5534 		ibdev_err(ibdev, "failed to reset SCC ctx, ret = %d.\n", ret);
5535 		goto out;
5536 	}
5537 
5538 	/* clear scc context */
5539 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CLR_SCCC, false);
5540 	clr = (struct hns_roce_sccc_clr *)desc.data;
5541 	clr->qpn = cpu_to_le32(hr_qp->qpn);
5542 	ret =  hns_roce_cmq_send(hr_dev, &desc, 1);
5543 	if (ret) {
5544 		ibdev_err(ibdev, "failed to clear SCC ctx, ret = %d.\n", ret);
5545 		goto out;
5546 	}
5547 
5548 	/* query scc context clear is done or not */
5549 	resp = (struct hns_roce_sccc_clr_done *)desc.data;
5550 	for (i = 0; i <= HNS_ROCE_CMQ_SCC_CLR_DONE_CNT; i++) {
5551 		hns_roce_cmq_setup_basic_desc(&desc,
5552 					      HNS_ROCE_OPC_QUERY_SCCC, true);
5553 		ret = hns_roce_cmq_send(hr_dev, &desc, 1);
5554 		if (ret) {
5555 			ibdev_err(ibdev, "failed to query clr cmq, ret = %d\n",
5556 				  ret);
5557 			goto out;
5558 		}
5559 
5560 		if (resp->clr_done)
5561 			goto out;
5562 
5563 		msleep(20);
5564 	}
5565 
5566 	ibdev_err(ibdev, "query SCC clr done flag overtime.\n");
5567 	ret = -ETIMEDOUT;
5568 
5569 out:
5570 	mutex_unlock(&hr_dev->qp_table.scc_mutex);
5571 	return ret;
5572 }
5573 
5574 #define DMA_IDX_SHIFT 3
5575 #define DMA_WQE_SHIFT 3
5576 
5577 static int hns_roce_v2_write_srqc_index_queue(struct hns_roce_srq *srq,
5578 					      struct hns_roce_srq_context *ctx)
5579 {
5580 	struct hns_roce_idx_que *idx_que = &srq->idx_que;
5581 	struct ib_device *ibdev = srq->ibsrq.device;
5582 	struct hns_roce_dev *hr_dev = to_hr_dev(ibdev);
5583 	u64 mtts_idx[MTT_MIN_COUNT] = {};
5584 	dma_addr_t dma_handle_idx = 0;
5585 	int ret;
5586 
5587 	/* Get physical address of idx que buf */
5588 	ret = hns_roce_mtr_find(hr_dev, &idx_que->mtr, 0, mtts_idx,
5589 				ARRAY_SIZE(mtts_idx), &dma_handle_idx);
5590 	if (ret < 1) {
5591 		ibdev_err(ibdev, "failed to find mtr for SRQ idx, ret = %d.\n",
5592 			  ret);
5593 		return -ENOBUFS;
5594 	}
5595 
5596 	hr_reg_write(ctx, SRQC_IDX_HOP_NUM,
5597 		     to_hr_hem_hopnum(hr_dev->caps.idx_hop_num, srq->wqe_cnt));
5598 
5599 	hr_reg_write(ctx, SRQC_IDX_BT_BA_L, dma_handle_idx >> DMA_IDX_SHIFT);
5600 	hr_reg_write(ctx, SRQC_IDX_BT_BA_H,
5601 		     upper_32_bits(dma_handle_idx >> DMA_IDX_SHIFT));
5602 
5603 	hr_reg_write(ctx, SRQC_IDX_BA_PG_SZ,
5604 		     to_hr_hw_page_shift(idx_que->mtr.hem_cfg.ba_pg_shift));
5605 	hr_reg_write(ctx, SRQC_IDX_BUF_PG_SZ,
5606 		     to_hr_hw_page_shift(idx_que->mtr.hem_cfg.buf_pg_shift));
5607 
5608 	hr_reg_write(ctx, SRQC_IDX_CUR_BLK_ADDR_L,
5609 		     to_hr_hw_page_addr(mtts_idx[0]));
5610 	hr_reg_write(ctx, SRQC_IDX_CUR_BLK_ADDR_H,
5611 		     upper_32_bits(to_hr_hw_page_addr(mtts_idx[0])));
5612 
5613 	hr_reg_write(ctx, SRQC_IDX_NXT_BLK_ADDR_L,
5614 		     to_hr_hw_page_addr(mtts_idx[1]));
5615 	hr_reg_write(ctx, SRQC_IDX_NXT_BLK_ADDR_H,
5616 		     upper_32_bits(to_hr_hw_page_addr(mtts_idx[1])));
5617 
5618 	return 0;
5619 }
5620 
5621 static int hns_roce_v2_write_srqc(struct hns_roce_srq *srq, void *mb_buf)
5622 {
5623 	struct ib_device *ibdev = srq->ibsrq.device;
5624 	struct hns_roce_dev *hr_dev = to_hr_dev(ibdev);
5625 	struct hns_roce_srq_context *ctx = mb_buf;
5626 	u64 mtts_wqe[MTT_MIN_COUNT] = {};
5627 	dma_addr_t dma_handle_wqe = 0;
5628 	int ret;
5629 
5630 	memset(ctx, 0, sizeof(*ctx));
5631 
5632 	/* Get the physical address of srq buf */
5633 	ret = hns_roce_mtr_find(hr_dev, &srq->buf_mtr, 0, mtts_wqe,
5634 				ARRAY_SIZE(mtts_wqe), &dma_handle_wqe);
5635 	if (ret < 1) {
5636 		ibdev_err(ibdev, "failed to find mtr for SRQ WQE, ret = %d.\n",
5637 			  ret);
5638 		return -ENOBUFS;
5639 	}
5640 
5641 	hr_reg_write(ctx, SRQC_SRQ_ST, 1);
5642 	hr_reg_write_bool(ctx, SRQC_SRQ_TYPE,
5643 			  srq->ibsrq.srq_type == IB_SRQT_XRC);
5644 	hr_reg_write(ctx, SRQC_PD, to_hr_pd(srq->ibsrq.pd)->pdn);
5645 	hr_reg_write(ctx, SRQC_SRQN, srq->srqn);
5646 	hr_reg_write(ctx, SRQC_XRCD, srq->xrcdn);
5647 	hr_reg_write(ctx, SRQC_XRC_CQN, srq->cqn);
5648 	hr_reg_write(ctx, SRQC_SHIFT, ilog2(srq->wqe_cnt));
5649 	hr_reg_write(ctx, SRQC_RQWS,
5650 		     srq->max_gs <= 0 ? 0 : fls(srq->max_gs - 1));
5651 
5652 	hr_reg_write(ctx, SRQC_WQE_HOP_NUM,
5653 		     to_hr_hem_hopnum(hr_dev->caps.srqwqe_hop_num,
5654 				      srq->wqe_cnt));
5655 
5656 	hr_reg_write(ctx, SRQC_WQE_BT_BA_L, dma_handle_wqe >> DMA_WQE_SHIFT);
5657 	hr_reg_write(ctx, SRQC_WQE_BT_BA_H,
5658 		     upper_32_bits(dma_handle_wqe >> DMA_WQE_SHIFT));
5659 
5660 	hr_reg_write(ctx, SRQC_WQE_BA_PG_SZ,
5661 		     to_hr_hw_page_shift(srq->buf_mtr.hem_cfg.ba_pg_shift));
5662 	hr_reg_write(ctx, SRQC_WQE_BUF_PG_SZ,
5663 		     to_hr_hw_page_shift(srq->buf_mtr.hem_cfg.buf_pg_shift));
5664 
5665 	if (srq->cap_flags & HNS_ROCE_SRQ_CAP_RECORD_DB) {
5666 		hr_reg_enable(ctx, SRQC_DB_RECORD_EN);
5667 		hr_reg_write(ctx, SRQC_DB_RECORD_ADDR_L,
5668 			     lower_32_bits(srq->rdb.dma) >> 1);
5669 		hr_reg_write(ctx, SRQC_DB_RECORD_ADDR_H,
5670 			     upper_32_bits(srq->rdb.dma));
5671 	}
5672 
5673 	return hns_roce_v2_write_srqc_index_queue(srq, ctx);
5674 }
5675 
5676 static int hns_roce_v2_modify_srq(struct ib_srq *ibsrq,
5677 				  struct ib_srq_attr *srq_attr,
5678 				  enum ib_srq_attr_mask srq_attr_mask,
5679 				  struct ib_udata *udata)
5680 {
5681 	struct hns_roce_dev *hr_dev = to_hr_dev(ibsrq->device);
5682 	struct hns_roce_srq *srq = to_hr_srq(ibsrq);
5683 	struct hns_roce_srq_context *srq_context;
5684 	struct hns_roce_srq_context *srqc_mask;
5685 	struct hns_roce_cmd_mailbox *mailbox;
5686 	int ret = 0;
5687 
5688 	/* Resizing SRQs is not supported yet */
5689 	if (srq_attr_mask & IB_SRQ_MAX_WR) {
5690 		ret = -EOPNOTSUPP;
5691 		goto out;
5692 	}
5693 
5694 	if (srq_attr_mask & IB_SRQ_LIMIT) {
5695 		if (srq_attr->srq_limit > srq->wqe_cnt) {
5696 			ret = -EINVAL;
5697 			goto out;
5698 		}
5699 
5700 		mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5701 		if (IS_ERR(mailbox)) {
5702 			ret = PTR_ERR(mailbox);
5703 			goto out;
5704 		}
5705 
5706 		srq_context = mailbox->buf;
5707 		srqc_mask = (struct hns_roce_srq_context *)mailbox->buf + 1;
5708 
5709 		memset(srqc_mask, 0xff, sizeof(*srqc_mask));
5710 
5711 		hr_reg_write(srq_context, SRQC_LIMIT_WL, srq_attr->srq_limit);
5712 		hr_reg_clear(srqc_mask, SRQC_LIMIT_WL);
5713 
5714 		ret = hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0,
5715 					HNS_ROCE_CMD_MODIFY_SRQC, srq->srqn);
5716 		hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5717 		if (ret)
5718 			ibdev_err(&hr_dev->ib_dev,
5719 				  "failed to handle cmd of modifying SRQ, ret = %d.\n",
5720 				  ret);
5721 	}
5722 
5723 out:
5724 	if (ret)
5725 		atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_SRQ_MODIFY_ERR_CNT]);
5726 
5727 	return ret;
5728 }
5729 
5730 static int hns_roce_v2_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr)
5731 {
5732 	struct hns_roce_dev *hr_dev = to_hr_dev(ibsrq->device);
5733 	struct hns_roce_srq *srq = to_hr_srq(ibsrq);
5734 	struct hns_roce_srq_context *srq_context;
5735 	struct hns_roce_cmd_mailbox *mailbox;
5736 	int ret;
5737 
5738 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5739 	if (IS_ERR(mailbox))
5740 		return PTR_ERR(mailbox);
5741 
5742 	srq_context = mailbox->buf;
5743 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma,
5744 				HNS_ROCE_CMD_QUERY_SRQC, srq->srqn);
5745 	if (ret) {
5746 		ibdev_err(&hr_dev->ib_dev,
5747 			  "failed to process cmd of querying SRQ, ret = %d.\n",
5748 			  ret);
5749 		goto out;
5750 	}
5751 
5752 	attr->srq_limit = hr_reg_read(srq_context, SRQC_LIMIT_WL);
5753 	attr->max_wr = srq->wqe_cnt;
5754 	attr->max_sge = srq->max_gs - srq->rsv_sge;
5755 
5756 out:
5757 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5758 	return ret;
5759 }
5760 
5761 static int hns_roce_v2_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
5762 {
5763 	struct hns_roce_dev *hr_dev = to_hr_dev(cq->device);
5764 	struct hns_roce_v2_cq_context *cq_context;
5765 	struct hns_roce_cq *hr_cq = to_hr_cq(cq);
5766 	struct hns_roce_v2_cq_context *cqc_mask;
5767 	struct hns_roce_cmd_mailbox *mailbox;
5768 	int ret;
5769 
5770 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5771 	ret = PTR_ERR_OR_ZERO(mailbox);
5772 	if (ret)
5773 		goto err_out;
5774 
5775 	cq_context = mailbox->buf;
5776 	cqc_mask = (struct hns_roce_v2_cq_context *)mailbox->buf + 1;
5777 
5778 	memset(cqc_mask, 0xff, sizeof(*cqc_mask));
5779 
5780 	hr_reg_write(cq_context, CQC_CQ_MAX_CNT, cq_count);
5781 	hr_reg_clear(cqc_mask, CQC_CQ_MAX_CNT);
5782 
5783 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
5784 		if (cq_period * HNS_ROCE_CLOCK_ADJUST > USHRT_MAX) {
5785 			dev_info(hr_dev->dev,
5786 				 "cq_period(%u) reached the upper limit, adjusted to 65.\n",
5787 				 cq_period);
5788 			cq_period = HNS_ROCE_MAX_CQ_PERIOD;
5789 		}
5790 		cq_period *= HNS_ROCE_CLOCK_ADJUST;
5791 	}
5792 	hr_reg_write(cq_context, CQC_CQ_PERIOD, cq_period);
5793 	hr_reg_clear(cqc_mask, CQC_CQ_PERIOD);
5794 
5795 	ret = hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0,
5796 				HNS_ROCE_CMD_MODIFY_CQC, hr_cq->cqn);
5797 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5798 	if (ret)
5799 		ibdev_err(&hr_dev->ib_dev,
5800 			  "failed to process cmd when modifying CQ, ret = %d.\n",
5801 			  ret);
5802 
5803 err_out:
5804 	if (ret)
5805 		atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_CQ_MODIFY_ERR_CNT]);
5806 
5807 	return ret;
5808 }
5809 
5810 static int hns_roce_v2_query_cqc(struct hns_roce_dev *hr_dev, u32 cqn,
5811 				 void *buffer)
5812 {
5813 	struct hns_roce_v2_cq_context *context;
5814 	struct hns_roce_cmd_mailbox *mailbox;
5815 	int ret;
5816 
5817 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5818 	if (IS_ERR(mailbox))
5819 		return PTR_ERR(mailbox);
5820 
5821 	context = mailbox->buf;
5822 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma,
5823 				HNS_ROCE_CMD_QUERY_CQC, cqn);
5824 	if (ret) {
5825 		ibdev_err(&hr_dev->ib_dev,
5826 			  "failed to process cmd when querying CQ, ret = %d.\n",
5827 			  ret);
5828 		goto err_mailbox;
5829 	}
5830 
5831 	memcpy(buffer, context, sizeof(*context));
5832 
5833 err_mailbox:
5834 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5835 
5836 	return ret;
5837 }
5838 
5839 static int hns_roce_v2_query_mpt(struct hns_roce_dev *hr_dev, u32 key,
5840 				 void *buffer)
5841 {
5842 	struct hns_roce_v2_mpt_entry *context;
5843 	struct hns_roce_cmd_mailbox *mailbox;
5844 	int ret;
5845 
5846 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5847 	if (IS_ERR(mailbox))
5848 		return PTR_ERR(mailbox);
5849 
5850 	context = mailbox->buf;
5851 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, HNS_ROCE_CMD_QUERY_MPT,
5852 				key_to_hw_index(key));
5853 	if (ret) {
5854 		ibdev_err(&hr_dev->ib_dev,
5855 			  "failed to process cmd when querying MPT, ret = %d.\n",
5856 			  ret);
5857 		goto err_mailbox;
5858 	}
5859 
5860 	memcpy(buffer, context, sizeof(*context));
5861 
5862 err_mailbox:
5863 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5864 
5865 	return ret;
5866 }
5867 
5868 static void hns_roce_irq_work_handle(struct work_struct *work)
5869 {
5870 	struct hns_roce_work *irq_work =
5871 				container_of(work, struct hns_roce_work, work);
5872 	struct ib_device *ibdev = &irq_work->hr_dev->ib_dev;
5873 
5874 	switch (irq_work->event_type) {
5875 	case HNS_ROCE_EVENT_TYPE_PATH_MIG:
5876 		ibdev_info(ibdev, "path migrated succeeded.\n");
5877 		break;
5878 	case HNS_ROCE_EVENT_TYPE_PATH_MIG_FAILED:
5879 		ibdev_warn(ibdev, "path migration failed.\n");
5880 		break;
5881 	case HNS_ROCE_EVENT_TYPE_COMM_EST:
5882 		break;
5883 	case HNS_ROCE_EVENT_TYPE_SQ_DRAINED:
5884 		ibdev_dbg(ibdev, "send queue drained.\n");
5885 		break;
5886 	case HNS_ROCE_EVENT_TYPE_WQ_CATAS_ERROR:
5887 		ibdev_err(ibdev, "local work queue 0x%x catast error, sub_event type is: %d\n",
5888 			  irq_work->queue_num, irq_work->sub_type);
5889 		break;
5890 	case HNS_ROCE_EVENT_TYPE_INV_REQ_LOCAL_WQ_ERROR:
5891 		ibdev_err(ibdev, "invalid request local work queue 0x%x error.\n",
5892 			  irq_work->queue_num);
5893 		break;
5894 	case HNS_ROCE_EVENT_TYPE_LOCAL_WQ_ACCESS_ERROR:
5895 		ibdev_err(ibdev, "local access violation work queue 0x%x error, sub_event type is: %d\n",
5896 			  irq_work->queue_num, irq_work->sub_type);
5897 		break;
5898 	case HNS_ROCE_EVENT_TYPE_SRQ_LIMIT_REACH:
5899 		ibdev_dbg(ibdev, "SRQ limit reach.\n");
5900 		break;
5901 	case HNS_ROCE_EVENT_TYPE_SRQ_LAST_WQE_REACH:
5902 		ibdev_dbg(ibdev, "SRQ last wqe reach.\n");
5903 		break;
5904 	case HNS_ROCE_EVENT_TYPE_SRQ_CATAS_ERROR:
5905 		ibdev_err(ibdev, "SRQ catas error.\n");
5906 		break;
5907 	case HNS_ROCE_EVENT_TYPE_CQ_ACCESS_ERROR:
5908 		ibdev_err(ibdev, "CQ 0x%x access err.\n", irq_work->queue_num);
5909 		break;
5910 	case HNS_ROCE_EVENT_TYPE_CQ_OVERFLOW:
5911 		ibdev_warn(ibdev, "CQ 0x%x overflow\n", irq_work->queue_num);
5912 		break;
5913 	case HNS_ROCE_EVENT_TYPE_DB_OVERFLOW:
5914 		ibdev_warn(ibdev, "DB overflow.\n");
5915 		break;
5916 	case HNS_ROCE_EVENT_TYPE_FLR:
5917 		ibdev_warn(ibdev, "function level reset.\n");
5918 		break;
5919 	case HNS_ROCE_EVENT_TYPE_XRCD_VIOLATION:
5920 		ibdev_err(ibdev, "xrc domain violation error.\n");
5921 		break;
5922 	case HNS_ROCE_EVENT_TYPE_INVALID_XRCETH:
5923 		ibdev_err(ibdev, "invalid xrceth error.\n");
5924 		break;
5925 	default:
5926 		break;
5927 	}
5928 
5929 	kfree(irq_work);
5930 }
5931 
5932 static void hns_roce_v2_init_irq_work(struct hns_roce_dev *hr_dev,
5933 				      struct hns_roce_eq *eq, u32 queue_num)
5934 {
5935 	struct hns_roce_work *irq_work;
5936 
5937 	irq_work = kzalloc(sizeof(struct hns_roce_work), GFP_ATOMIC);
5938 	if (!irq_work)
5939 		return;
5940 
5941 	INIT_WORK(&irq_work->work, hns_roce_irq_work_handle);
5942 	irq_work->hr_dev = hr_dev;
5943 	irq_work->event_type = eq->event_type;
5944 	irq_work->sub_type = eq->sub_type;
5945 	irq_work->queue_num = queue_num;
5946 	queue_work(hr_dev->irq_workq, &irq_work->work);
5947 }
5948 
5949 static void update_eq_db(struct hns_roce_eq *eq)
5950 {
5951 	struct hns_roce_dev *hr_dev = eq->hr_dev;
5952 	struct hns_roce_v2_db eq_db = {};
5953 
5954 	if (eq->type_flag == HNS_ROCE_AEQ) {
5955 		hr_reg_write(&eq_db, EQ_DB_CMD,
5956 			     eq->arm_st == HNS_ROCE_V2_EQ_ALWAYS_ARMED ?
5957 			     HNS_ROCE_EQ_DB_CMD_AEQ :
5958 			     HNS_ROCE_EQ_DB_CMD_AEQ_ARMED);
5959 	} else {
5960 		hr_reg_write(&eq_db, EQ_DB_TAG, eq->eqn);
5961 
5962 		hr_reg_write(&eq_db, EQ_DB_CMD,
5963 			     eq->arm_st == HNS_ROCE_V2_EQ_ALWAYS_ARMED ?
5964 			     HNS_ROCE_EQ_DB_CMD_CEQ :
5965 			     HNS_ROCE_EQ_DB_CMD_CEQ_ARMED);
5966 	}
5967 
5968 	hr_reg_write(&eq_db, EQ_DB_CI, eq->cons_index);
5969 
5970 	hns_roce_write64(hr_dev, (__le32 *)&eq_db, eq->db_reg);
5971 }
5972 
5973 static struct hns_roce_aeqe *next_aeqe_sw_v2(struct hns_roce_eq *eq)
5974 {
5975 	struct hns_roce_aeqe *aeqe;
5976 
5977 	aeqe = hns_roce_buf_offset(eq->mtr.kmem,
5978 				   (eq->cons_index & (eq->entries - 1)) *
5979 				   eq->eqe_size);
5980 
5981 	return (hr_reg_read(aeqe, AEQE_OWNER) ^
5982 		!!(eq->cons_index & eq->entries)) ? aeqe : NULL;
5983 }
5984 
5985 static irqreturn_t hns_roce_v2_aeq_int(struct hns_roce_dev *hr_dev,
5986 				       struct hns_roce_eq *eq)
5987 {
5988 	struct device *dev = hr_dev->dev;
5989 	struct hns_roce_aeqe *aeqe = next_aeqe_sw_v2(eq);
5990 	irqreturn_t aeqe_found = IRQ_NONE;
5991 	int event_type;
5992 	u32 queue_num;
5993 	int sub_type;
5994 
5995 	while (aeqe) {
5996 		/* Make sure we read AEQ entry after we have checked the
5997 		 * ownership bit
5998 		 */
5999 		dma_rmb();
6000 
6001 		event_type = hr_reg_read(aeqe, AEQE_EVENT_TYPE);
6002 		sub_type = hr_reg_read(aeqe, AEQE_SUB_TYPE);
6003 		queue_num = hr_reg_read(aeqe, AEQE_EVENT_QUEUE_NUM);
6004 
6005 		switch (event_type) {
6006 		case HNS_ROCE_EVENT_TYPE_PATH_MIG:
6007 		case HNS_ROCE_EVENT_TYPE_PATH_MIG_FAILED:
6008 		case HNS_ROCE_EVENT_TYPE_COMM_EST:
6009 		case HNS_ROCE_EVENT_TYPE_SQ_DRAINED:
6010 		case HNS_ROCE_EVENT_TYPE_WQ_CATAS_ERROR:
6011 		case HNS_ROCE_EVENT_TYPE_SRQ_LAST_WQE_REACH:
6012 		case HNS_ROCE_EVENT_TYPE_INV_REQ_LOCAL_WQ_ERROR:
6013 		case HNS_ROCE_EVENT_TYPE_LOCAL_WQ_ACCESS_ERROR:
6014 		case HNS_ROCE_EVENT_TYPE_XRCD_VIOLATION:
6015 		case HNS_ROCE_EVENT_TYPE_INVALID_XRCETH:
6016 			hns_roce_qp_event(hr_dev, queue_num, event_type);
6017 			break;
6018 		case HNS_ROCE_EVENT_TYPE_SRQ_LIMIT_REACH:
6019 		case HNS_ROCE_EVENT_TYPE_SRQ_CATAS_ERROR:
6020 			hns_roce_srq_event(hr_dev, queue_num, event_type);
6021 			break;
6022 		case HNS_ROCE_EVENT_TYPE_CQ_ACCESS_ERROR:
6023 		case HNS_ROCE_EVENT_TYPE_CQ_OVERFLOW:
6024 			hns_roce_cq_event(hr_dev, queue_num, event_type);
6025 			break;
6026 		case HNS_ROCE_EVENT_TYPE_MB:
6027 			hns_roce_cmd_event(hr_dev,
6028 					le16_to_cpu(aeqe->event.cmd.token),
6029 					aeqe->event.cmd.status,
6030 					le64_to_cpu(aeqe->event.cmd.out_param));
6031 			break;
6032 		case HNS_ROCE_EVENT_TYPE_DB_OVERFLOW:
6033 		case HNS_ROCE_EVENT_TYPE_FLR:
6034 			break;
6035 		default:
6036 			dev_err(dev, "unhandled event %d on EQ %d at idx %u.\n",
6037 				event_type, eq->eqn, eq->cons_index);
6038 			break;
6039 		}
6040 
6041 		eq->event_type = event_type;
6042 		eq->sub_type = sub_type;
6043 		++eq->cons_index;
6044 		aeqe_found = IRQ_HANDLED;
6045 
6046 		atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_AEQE_CNT]);
6047 
6048 		hns_roce_v2_init_irq_work(hr_dev, eq, queue_num);
6049 
6050 		aeqe = next_aeqe_sw_v2(eq);
6051 	}
6052 
6053 	update_eq_db(eq);
6054 
6055 	return IRQ_RETVAL(aeqe_found);
6056 }
6057 
6058 static struct hns_roce_ceqe *next_ceqe_sw_v2(struct hns_roce_eq *eq)
6059 {
6060 	struct hns_roce_ceqe *ceqe;
6061 
6062 	ceqe = hns_roce_buf_offset(eq->mtr.kmem,
6063 				   (eq->cons_index & (eq->entries - 1)) *
6064 				   eq->eqe_size);
6065 
6066 	return (hr_reg_read(ceqe, CEQE_OWNER) ^
6067 		!!(eq->cons_index & eq->entries)) ? ceqe : NULL;
6068 }
6069 
6070 static irqreturn_t hns_roce_v2_ceq_int(struct hns_roce_dev *hr_dev,
6071 				       struct hns_roce_eq *eq)
6072 {
6073 	struct hns_roce_ceqe *ceqe = next_ceqe_sw_v2(eq);
6074 	irqreturn_t ceqe_found = IRQ_NONE;
6075 	u32 cqn;
6076 
6077 	while (ceqe) {
6078 		/* Make sure we read CEQ entry after we have checked the
6079 		 * ownership bit
6080 		 */
6081 		dma_rmb();
6082 
6083 		cqn = hr_reg_read(ceqe, CEQE_CQN);
6084 
6085 		hns_roce_cq_completion(hr_dev, cqn);
6086 
6087 		++eq->cons_index;
6088 		ceqe_found = IRQ_HANDLED;
6089 		atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_CEQE_CNT]);
6090 
6091 		ceqe = next_ceqe_sw_v2(eq);
6092 	}
6093 
6094 	update_eq_db(eq);
6095 
6096 	return IRQ_RETVAL(ceqe_found);
6097 }
6098 
6099 static irqreturn_t hns_roce_v2_msix_interrupt_eq(int irq, void *eq_ptr)
6100 {
6101 	struct hns_roce_eq *eq = eq_ptr;
6102 	struct hns_roce_dev *hr_dev = eq->hr_dev;
6103 	irqreturn_t int_work;
6104 
6105 	if (eq->type_flag == HNS_ROCE_CEQ)
6106 		/* Completion event interrupt */
6107 		int_work = hns_roce_v2_ceq_int(hr_dev, eq);
6108 	else
6109 		/* Asynchronous event interrupt */
6110 		int_work = hns_roce_v2_aeq_int(hr_dev, eq);
6111 
6112 	return IRQ_RETVAL(int_work);
6113 }
6114 
6115 static irqreturn_t abnormal_interrupt_basic(struct hns_roce_dev *hr_dev,
6116 					    u32 int_st)
6117 {
6118 	struct pci_dev *pdev = hr_dev->pci_dev;
6119 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
6120 	const struct hnae3_ae_ops *ops = ae_dev->ops;
6121 	irqreturn_t int_work = IRQ_NONE;
6122 	u32 int_en;
6123 
6124 	int_en = roce_read(hr_dev, ROCEE_VF_ABN_INT_EN_REG);
6125 
6126 	if (int_st & BIT(HNS_ROCE_V2_VF_INT_ST_AEQ_OVERFLOW_S)) {
6127 		dev_err(hr_dev->dev, "AEQ overflow!\n");
6128 
6129 		roce_write(hr_dev, ROCEE_VF_ABN_INT_ST_REG,
6130 			   1 << HNS_ROCE_V2_VF_INT_ST_AEQ_OVERFLOW_S);
6131 
6132 		/* Set reset level for reset_event() */
6133 		if (ops->set_default_reset_request)
6134 			ops->set_default_reset_request(ae_dev,
6135 						       HNAE3_FUNC_RESET);
6136 		if (ops->reset_event)
6137 			ops->reset_event(pdev, NULL);
6138 
6139 		int_en |= 1 << HNS_ROCE_V2_VF_ABN_INT_EN_S;
6140 		roce_write(hr_dev, ROCEE_VF_ABN_INT_EN_REG, int_en);
6141 
6142 		int_work = IRQ_HANDLED;
6143 	} else {
6144 		dev_err(hr_dev->dev, "there is no basic abn irq found.\n");
6145 	}
6146 
6147 	return IRQ_RETVAL(int_work);
6148 }
6149 
6150 static int fmea_ram_ecc_query(struct hns_roce_dev *hr_dev,
6151 			       struct fmea_ram_ecc *ecc_info)
6152 {
6153 	struct hns_roce_cmq_desc desc;
6154 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
6155 	int ret;
6156 
6157 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_QUERY_RAM_ECC, true);
6158 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
6159 	if (ret)
6160 		return ret;
6161 
6162 	ecc_info->is_ecc_err = hr_reg_read(req, QUERY_RAM_ECC_1BIT_ERR);
6163 	ecc_info->res_type = hr_reg_read(req, QUERY_RAM_ECC_RES_TYPE);
6164 	ecc_info->index = hr_reg_read(req, QUERY_RAM_ECC_TAG);
6165 
6166 	return 0;
6167 }
6168 
6169 static int fmea_recover_gmv(struct hns_roce_dev *hr_dev, u32 idx)
6170 {
6171 	struct hns_roce_cmq_desc desc;
6172 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
6173 	u32 addr_upper;
6174 	u32 addr_low;
6175 	int ret;
6176 
6177 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_GMV_BT, true);
6178 	hr_reg_write(req, CFG_GMV_BT_IDX, idx);
6179 
6180 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
6181 	if (ret) {
6182 		dev_err(hr_dev->dev,
6183 			"failed to execute cmd to read gmv, ret = %d.\n", ret);
6184 		return ret;
6185 	}
6186 
6187 	addr_low =  hr_reg_read(req, CFG_GMV_BT_BA_L);
6188 	addr_upper = hr_reg_read(req, CFG_GMV_BT_BA_H);
6189 
6190 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_GMV_BT, false);
6191 	hr_reg_write(req, CFG_GMV_BT_BA_L, addr_low);
6192 	hr_reg_write(req, CFG_GMV_BT_BA_H, addr_upper);
6193 	hr_reg_write(req, CFG_GMV_BT_IDX, idx);
6194 
6195 	return hns_roce_cmq_send(hr_dev, &desc, 1);
6196 }
6197 
6198 static u64 fmea_get_ram_res_addr(u32 res_type, __le64 *data)
6199 {
6200 	if (res_type == ECC_RESOURCE_QPC_TIMER ||
6201 	    res_type == ECC_RESOURCE_CQC_TIMER ||
6202 	    res_type == ECC_RESOURCE_SCCC)
6203 		return le64_to_cpu(*data);
6204 
6205 	return le64_to_cpu(*data) << PAGE_SHIFT;
6206 }
6207 
6208 static int fmea_recover_others(struct hns_roce_dev *hr_dev, u32 res_type,
6209 			       u32 index)
6210 {
6211 	u8 write_bt0_op = fmea_ram_res[res_type].write_bt0_op;
6212 	u8 read_bt0_op = fmea_ram_res[res_type].read_bt0_op;
6213 	struct hns_roce_cmd_mailbox *mailbox;
6214 	u64 addr;
6215 	int ret;
6216 
6217 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
6218 	if (IS_ERR(mailbox))
6219 		return PTR_ERR(mailbox);
6220 
6221 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, read_bt0_op, index);
6222 	if (ret) {
6223 		dev_err(hr_dev->dev,
6224 			"failed to execute cmd to read fmea ram, ret = %d.\n",
6225 			ret);
6226 		goto out;
6227 	}
6228 
6229 	addr = fmea_get_ram_res_addr(res_type, mailbox->buf);
6230 
6231 	ret = hns_roce_cmd_mbox(hr_dev, addr, 0, write_bt0_op, index);
6232 	if (ret)
6233 		dev_err(hr_dev->dev,
6234 			"failed to execute cmd to write fmea ram, ret = %d.\n",
6235 			ret);
6236 
6237 out:
6238 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
6239 	return ret;
6240 }
6241 
6242 static void fmea_ram_ecc_recover(struct hns_roce_dev *hr_dev,
6243 				 struct fmea_ram_ecc *ecc_info)
6244 {
6245 	u32 res_type = ecc_info->res_type;
6246 	u32 index = ecc_info->index;
6247 	int ret;
6248 
6249 	BUILD_BUG_ON(ARRAY_SIZE(fmea_ram_res) != ECC_RESOURCE_COUNT);
6250 
6251 	if (res_type >= ECC_RESOURCE_COUNT) {
6252 		dev_err(hr_dev->dev, "unsupported fmea ram ecc type %u.\n",
6253 			res_type);
6254 		return;
6255 	}
6256 
6257 	if (res_type == ECC_RESOURCE_GMV)
6258 		ret = fmea_recover_gmv(hr_dev, index);
6259 	else
6260 		ret = fmea_recover_others(hr_dev, res_type, index);
6261 	if (ret)
6262 		dev_err(hr_dev->dev,
6263 			"failed to recover %s, index = %u, ret = %d.\n",
6264 			fmea_ram_res[res_type].name, index, ret);
6265 }
6266 
6267 static void fmea_ram_ecc_work(struct work_struct *ecc_work)
6268 {
6269 	struct hns_roce_dev *hr_dev =
6270 		container_of(ecc_work, struct hns_roce_dev, ecc_work);
6271 	struct fmea_ram_ecc ecc_info = {};
6272 
6273 	if (fmea_ram_ecc_query(hr_dev, &ecc_info)) {
6274 		dev_err(hr_dev->dev, "failed to query fmea ram ecc.\n");
6275 		return;
6276 	}
6277 
6278 	if (!ecc_info.is_ecc_err) {
6279 		dev_err(hr_dev->dev, "there is no fmea ram ecc err found.\n");
6280 		return;
6281 	}
6282 
6283 	fmea_ram_ecc_recover(hr_dev, &ecc_info);
6284 }
6285 
6286 static irqreturn_t hns_roce_v2_msix_interrupt_abn(int irq, void *dev_id)
6287 {
6288 	struct hns_roce_dev *hr_dev = dev_id;
6289 	irqreturn_t int_work = IRQ_NONE;
6290 	u32 int_st;
6291 
6292 	int_st = roce_read(hr_dev, ROCEE_VF_ABN_INT_ST_REG);
6293 
6294 	if (int_st) {
6295 		int_work = abnormal_interrupt_basic(hr_dev, int_st);
6296 	} else if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
6297 		queue_work(hr_dev->irq_workq, &hr_dev->ecc_work);
6298 		int_work = IRQ_HANDLED;
6299 	} else {
6300 		dev_err(hr_dev->dev, "there is no abnormal irq found.\n");
6301 	}
6302 
6303 	return IRQ_RETVAL(int_work);
6304 }
6305 
6306 static void hns_roce_v2_int_mask_enable(struct hns_roce_dev *hr_dev,
6307 					int eq_num, u32 enable_flag)
6308 {
6309 	int i;
6310 
6311 	for (i = 0; i < eq_num; i++)
6312 		roce_write(hr_dev, ROCEE_VF_EVENT_INT_EN_REG +
6313 			   i * EQ_REG_OFFSET, enable_flag);
6314 
6315 	roce_write(hr_dev, ROCEE_VF_ABN_INT_EN_REG, enable_flag);
6316 	roce_write(hr_dev, ROCEE_VF_ABN_INT_CFG_REG, enable_flag);
6317 }
6318 
6319 static void hns_roce_v2_destroy_eqc(struct hns_roce_dev *hr_dev, u32 eqn)
6320 {
6321 	struct device *dev = hr_dev->dev;
6322 	int ret;
6323 	u8 cmd;
6324 
6325 	if (eqn < hr_dev->caps.num_comp_vectors)
6326 		cmd = HNS_ROCE_CMD_DESTROY_CEQC;
6327 	else
6328 		cmd = HNS_ROCE_CMD_DESTROY_AEQC;
6329 
6330 	ret = hns_roce_destroy_hw_ctx(hr_dev, cmd, eqn & HNS_ROCE_V2_EQN_M);
6331 	if (ret)
6332 		dev_err(dev, "[mailbox cmd] destroy eqc(%u) failed.\n", eqn);
6333 }
6334 
6335 static void free_eq_buf(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq)
6336 {
6337 	hns_roce_mtr_destroy(hr_dev, &eq->mtr);
6338 }
6339 
6340 static void init_eq_config(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq)
6341 {
6342 	eq->db_reg = hr_dev->reg_base + ROCEE_VF_EQ_DB_CFG0_REG;
6343 	eq->cons_index = 0;
6344 	eq->over_ignore = HNS_ROCE_V2_EQ_OVER_IGNORE_0;
6345 	eq->coalesce = HNS_ROCE_V2_EQ_COALESCE_0;
6346 	eq->arm_st = HNS_ROCE_V2_EQ_ALWAYS_ARMED;
6347 	eq->shift = ilog2((unsigned int)eq->entries);
6348 }
6349 
6350 static int config_eqc(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq,
6351 		      void *mb_buf)
6352 {
6353 	u64 eqe_ba[MTT_MIN_COUNT] = { 0 };
6354 	struct hns_roce_eq_context *eqc;
6355 	u64 bt_ba = 0;
6356 	int count;
6357 
6358 	eqc = mb_buf;
6359 	memset(eqc, 0, sizeof(struct hns_roce_eq_context));
6360 
6361 	init_eq_config(hr_dev, eq);
6362 
6363 	/* if not multi-hop, eqe buffer only use one trunk */
6364 	count = hns_roce_mtr_find(hr_dev, &eq->mtr, 0, eqe_ba, MTT_MIN_COUNT,
6365 				  &bt_ba);
6366 	if (count < 1) {
6367 		dev_err(hr_dev->dev, "failed to find EQE mtr\n");
6368 		return -ENOBUFS;
6369 	}
6370 
6371 	hr_reg_write(eqc, EQC_EQ_ST, HNS_ROCE_V2_EQ_STATE_VALID);
6372 	hr_reg_write(eqc, EQC_EQE_HOP_NUM, eq->hop_num);
6373 	hr_reg_write(eqc, EQC_OVER_IGNORE, eq->over_ignore);
6374 	hr_reg_write(eqc, EQC_COALESCE, eq->coalesce);
6375 	hr_reg_write(eqc, EQC_ARM_ST, eq->arm_st);
6376 	hr_reg_write(eqc, EQC_EQN, eq->eqn);
6377 	hr_reg_write(eqc, EQC_EQE_CNT, HNS_ROCE_EQ_INIT_EQE_CNT);
6378 	hr_reg_write(eqc, EQC_EQE_BA_PG_SZ,
6379 		     to_hr_hw_page_shift(eq->mtr.hem_cfg.ba_pg_shift));
6380 	hr_reg_write(eqc, EQC_EQE_BUF_PG_SZ,
6381 		     to_hr_hw_page_shift(eq->mtr.hem_cfg.buf_pg_shift));
6382 	hr_reg_write(eqc, EQC_EQ_PROD_INDX, HNS_ROCE_EQ_INIT_PROD_IDX);
6383 	hr_reg_write(eqc, EQC_EQ_MAX_CNT, eq->eq_max_cnt);
6384 
6385 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
6386 		if (eq->eq_period * HNS_ROCE_CLOCK_ADJUST > USHRT_MAX) {
6387 			dev_info(hr_dev->dev, "eq_period(%u) reached the upper limit, adjusted to 65.\n",
6388 				 eq->eq_period);
6389 			eq->eq_period = HNS_ROCE_MAX_EQ_PERIOD;
6390 		}
6391 		eq->eq_period *= HNS_ROCE_CLOCK_ADJUST;
6392 	}
6393 
6394 	hr_reg_write(eqc, EQC_EQ_PERIOD, eq->eq_period);
6395 	hr_reg_write(eqc, EQC_EQE_REPORT_TIMER, HNS_ROCE_EQ_INIT_REPORT_TIMER);
6396 	hr_reg_write(eqc, EQC_EQE_BA_L, bt_ba >> 3);
6397 	hr_reg_write(eqc, EQC_EQE_BA_H, bt_ba >> 35);
6398 	hr_reg_write(eqc, EQC_SHIFT, eq->shift);
6399 	hr_reg_write(eqc, EQC_MSI_INDX, HNS_ROCE_EQ_INIT_MSI_IDX);
6400 	hr_reg_write(eqc, EQC_CUR_EQE_BA_L, eqe_ba[0] >> 12);
6401 	hr_reg_write(eqc, EQC_CUR_EQE_BA_M, eqe_ba[0] >> 28);
6402 	hr_reg_write(eqc, EQC_CUR_EQE_BA_H, eqe_ba[0] >> 60);
6403 	hr_reg_write(eqc, EQC_EQ_CONS_INDX, HNS_ROCE_EQ_INIT_CONS_IDX);
6404 	hr_reg_write(eqc, EQC_NEX_EQE_BA_L, eqe_ba[1] >> 12);
6405 	hr_reg_write(eqc, EQC_NEX_EQE_BA_H, eqe_ba[1] >> 44);
6406 	hr_reg_write(eqc, EQC_EQE_SIZE, eq->eqe_size == HNS_ROCE_V3_EQE_SIZE);
6407 
6408 	return 0;
6409 }
6410 
6411 static int alloc_eq_buf(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq)
6412 {
6413 	struct hns_roce_buf_attr buf_attr = {};
6414 	int err;
6415 
6416 	if (hr_dev->caps.eqe_hop_num == HNS_ROCE_HOP_NUM_0)
6417 		eq->hop_num = 0;
6418 	else
6419 		eq->hop_num = hr_dev->caps.eqe_hop_num;
6420 
6421 	buf_attr.page_shift = hr_dev->caps.eqe_buf_pg_sz + PAGE_SHIFT;
6422 	buf_attr.region[0].size = eq->entries * eq->eqe_size;
6423 	buf_attr.region[0].hopnum = eq->hop_num;
6424 	buf_attr.region_count = 1;
6425 
6426 	err = hns_roce_mtr_create(hr_dev, &eq->mtr, &buf_attr,
6427 				  hr_dev->caps.eqe_ba_pg_sz + PAGE_SHIFT, NULL,
6428 				  0);
6429 	if (err)
6430 		dev_err(hr_dev->dev, "failed to alloc EQE mtr, err %d\n", err);
6431 
6432 	return err;
6433 }
6434 
6435 static int hns_roce_v2_create_eq(struct hns_roce_dev *hr_dev,
6436 				 struct hns_roce_eq *eq, u8 eq_cmd)
6437 {
6438 	struct hns_roce_cmd_mailbox *mailbox;
6439 	int ret;
6440 
6441 	/* Allocate mailbox memory */
6442 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
6443 	if (IS_ERR(mailbox))
6444 		return PTR_ERR(mailbox);
6445 
6446 	ret = alloc_eq_buf(hr_dev, eq);
6447 	if (ret)
6448 		goto free_cmd_mbox;
6449 
6450 	ret = config_eqc(hr_dev, eq, mailbox->buf);
6451 	if (ret)
6452 		goto err_cmd_mbox;
6453 
6454 	ret = hns_roce_create_hw_ctx(hr_dev, mailbox, eq_cmd, eq->eqn);
6455 	if (ret) {
6456 		dev_err(hr_dev->dev, "[mailbox cmd] create eqc failed.\n");
6457 		goto err_cmd_mbox;
6458 	}
6459 
6460 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
6461 
6462 	return 0;
6463 
6464 err_cmd_mbox:
6465 	free_eq_buf(hr_dev, eq);
6466 
6467 free_cmd_mbox:
6468 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
6469 
6470 	return ret;
6471 }
6472 
6473 static int __hns_roce_request_irq(struct hns_roce_dev *hr_dev, int irq_num,
6474 				  int comp_num, int aeq_num, int other_num)
6475 {
6476 	struct hns_roce_eq_table *eq_table = &hr_dev->eq_table;
6477 	int i, j;
6478 	int ret;
6479 
6480 	for (i = 0; i < irq_num; i++) {
6481 		hr_dev->irq_names[i] = kzalloc(HNS_ROCE_INT_NAME_LEN,
6482 					       GFP_KERNEL);
6483 		if (!hr_dev->irq_names[i]) {
6484 			ret = -ENOMEM;
6485 			goto err_kzalloc_failed;
6486 		}
6487 	}
6488 
6489 	/* irq contains: abnormal + AEQ + CEQ */
6490 	for (j = 0; j < other_num; j++)
6491 		snprintf((char *)hr_dev->irq_names[j], HNS_ROCE_INT_NAME_LEN,
6492 			 "hns-%s-abn-%d", pci_name(hr_dev->pci_dev), j);
6493 
6494 	for (j = other_num; j < (other_num + aeq_num); j++)
6495 		snprintf((char *)hr_dev->irq_names[j], HNS_ROCE_INT_NAME_LEN,
6496 			 "hns-%s-aeq-%d", pci_name(hr_dev->pci_dev), j - other_num);
6497 
6498 	for (j = (other_num + aeq_num); j < irq_num; j++)
6499 		snprintf((char *)hr_dev->irq_names[j], HNS_ROCE_INT_NAME_LEN,
6500 			 "hns-%s-ceq-%d", pci_name(hr_dev->pci_dev),
6501 			 j - other_num - aeq_num);
6502 
6503 	for (j = 0; j < irq_num; j++) {
6504 		if (j < other_num)
6505 			ret = request_irq(hr_dev->irq[j],
6506 					  hns_roce_v2_msix_interrupt_abn,
6507 					  0, hr_dev->irq_names[j], hr_dev);
6508 
6509 		else if (j < (other_num + comp_num))
6510 			ret = request_irq(eq_table->eq[j - other_num].irq,
6511 					  hns_roce_v2_msix_interrupt_eq,
6512 					  0, hr_dev->irq_names[j + aeq_num],
6513 					  &eq_table->eq[j - other_num]);
6514 		else
6515 			ret = request_irq(eq_table->eq[j - other_num].irq,
6516 					  hns_roce_v2_msix_interrupt_eq,
6517 					  0, hr_dev->irq_names[j - comp_num],
6518 					  &eq_table->eq[j - other_num]);
6519 		if (ret) {
6520 			dev_err(hr_dev->dev, "request irq error!\n");
6521 			goto err_request_failed;
6522 		}
6523 	}
6524 
6525 	return 0;
6526 
6527 err_request_failed:
6528 	for (j -= 1; j >= 0; j--)
6529 		if (j < other_num)
6530 			free_irq(hr_dev->irq[j], hr_dev);
6531 		else
6532 			free_irq(eq_table->eq[j - other_num].irq,
6533 				 &eq_table->eq[j - other_num]);
6534 
6535 err_kzalloc_failed:
6536 	for (i -= 1; i >= 0; i--)
6537 		kfree(hr_dev->irq_names[i]);
6538 
6539 	return ret;
6540 }
6541 
6542 static void __hns_roce_free_irq(struct hns_roce_dev *hr_dev)
6543 {
6544 	int irq_num;
6545 	int eq_num;
6546 	int i;
6547 
6548 	eq_num = hr_dev->caps.num_comp_vectors + hr_dev->caps.num_aeq_vectors;
6549 	irq_num = eq_num + hr_dev->caps.num_other_vectors;
6550 
6551 	for (i = 0; i < hr_dev->caps.num_other_vectors; i++)
6552 		free_irq(hr_dev->irq[i], hr_dev);
6553 
6554 	for (i = 0; i < eq_num; i++)
6555 		free_irq(hr_dev->eq_table.eq[i].irq, &hr_dev->eq_table.eq[i]);
6556 
6557 	for (i = 0; i < irq_num; i++)
6558 		kfree(hr_dev->irq_names[i]);
6559 }
6560 
6561 static int hns_roce_v2_init_eq_table(struct hns_roce_dev *hr_dev)
6562 {
6563 	struct hns_roce_eq_table *eq_table = &hr_dev->eq_table;
6564 	struct device *dev = hr_dev->dev;
6565 	struct hns_roce_eq *eq;
6566 	int other_num;
6567 	int comp_num;
6568 	int aeq_num;
6569 	int irq_num;
6570 	int eq_num;
6571 	u8 eq_cmd;
6572 	int ret;
6573 	int i;
6574 
6575 	other_num = hr_dev->caps.num_other_vectors;
6576 	comp_num = hr_dev->caps.num_comp_vectors;
6577 	aeq_num = hr_dev->caps.num_aeq_vectors;
6578 
6579 	eq_num = comp_num + aeq_num;
6580 	irq_num = eq_num + other_num;
6581 
6582 	eq_table->eq = kcalloc(eq_num, sizeof(*eq_table->eq), GFP_KERNEL);
6583 	if (!eq_table->eq)
6584 		return -ENOMEM;
6585 
6586 	/* create eq */
6587 	for (i = 0; i < eq_num; i++) {
6588 		eq = &eq_table->eq[i];
6589 		eq->hr_dev = hr_dev;
6590 		eq->eqn = i;
6591 		if (i < comp_num) {
6592 			/* CEQ */
6593 			eq_cmd = HNS_ROCE_CMD_CREATE_CEQC;
6594 			eq->type_flag = HNS_ROCE_CEQ;
6595 			eq->entries = hr_dev->caps.ceqe_depth;
6596 			eq->eqe_size = hr_dev->caps.ceqe_size;
6597 			eq->irq = hr_dev->irq[i + other_num + aeq_num];
6598 			eq->eq_max_cnt = HNS_ROCE_CEQ_DEFAULT_BURST_NUM;
6599 			eq->eq_period = HNS_ROCE_CEQ_DEFAULT_INTERVAL;
6600 		} else {
6601 			/* AEQ */
6602 			eq_cmd = HNS_ROCE_CMD_CREATE_AEQC;
6603 			eq->type_flag = HNS_ROCE_AEQ;
6604 			eq->entries = hr_dev->caps.aeqe_depth;
6605 			eq->eqe_size = hr_dev->caps.aeqe_size;
6606 			eq->irq = hr_dev->irq[i - comp_num + other_num];
6607 			eq->eq_max_cnt = HNS_ROCE_AEQ_DEFAULT_BURST_NUM;
6608 			eq->eq_period = HNS_ROCE_AEQ_DEFAULT_INTERVAL;
6609 		}
6610 
6611 		ret = hns_roce_v2_create_eq(hr_dev, eq, eq_cmd);
6612 		if (ret) {
6613 			dev_err(dev, "failed to create eq.\n");
6614 			goto err_create_eq_fail;
6615 		}
6616 	}
6617 
6618 	INIT_WORK(&hr_dev->ecc_work, fmea_ram_ecc_work);
6619 
6620 	hr_dev->irq_workq = alloc_ordered_workqueue("hns_roce_irq_workq", 0);
6621 	if (!hr_dev->irq_workq) {
6622 		dev_err(dev, "failed to create irq workqueue.\n");
6623 		ret = -ENOMEM;
6624 		goto err_create_eq_fail;
6625 	}
6626 
6627 	ret = __hns_roce_request_irq(hr_dev, irq_num, comp_num, aeq_num,
6628 				     other_num);
6629 	if (ret) {
6630 		dev_err(dev, "failed to request irq.\n");
6631 		goto err_request_irq_fail;
6632 	}
6633 
6634 	/* enable irq */
6635 	hns_roce_v2_int_mask_enable(hr_dev, eq_num, EQ_ENABLE);
6636 
6637 	return 0;
6638 
6639 err_request_irq_fail:
6640 	destroy_workqueue(hr_dev->irq_workq);
6641 
6642 err_create_eq_fail:
6643 	for (i -= 1; i >= 0; i--)
6644 		free_eq_buf(hr_dev, &eq_table->eq[i]);
6645 	kfree(eq_table->eq);
6646 
6647 	return ret;
6648 }
6649 
6650 static void hns_roce_v2_cleanup_eq_table(struct hns_roce_dev *hr_dev)
6651 {
6652 	struct hns_roce_eq_table *eq_table = &hr_dev->eq_table;
6653 	int eq_num;
6654 	int i;
6655 
6656 	eq_num = hr_dev->caps.num_comp_vectors + hr_dev->caps.num_aeq_vectors;
6657 
6658 	/* Disable irq */
6659 	hns_roce_v2_int_mask_enable(hr_dev, eq_num, EQ_DISABLE);
6660 
6661 	__hns_roce_free_irq(hr_dev);
6662 	destroy_workqueue(hr_dev->irq_workq);
6663 
6664 	for (i = 0; i < eq_num; i++) {
6665 		hns_roce_v2_destroy_eqc(hr_dev, i);
6666 
6667 		free_eq_buf(hr_dev, &eq_table->eq[i]);
6668 	}
6669 
6670 	kfree(eq_table->eq);
6671 }
6672 
6673 static const struct ib_device_ops hns_roce_v2_dev_ops = {
6674 	.destroy_qp = hns_roce_v2_destroy_qp,
6675 	.modify_cq = hns_roce_v2_modify_cq,
6676 	.poll_cq = hns_roce_v2_poll_cq,
6677 	.post_recv = hns_roce_v2_post_recv,
6678 	.post_send = hns_roce_v2_post_send,
6679 	.query_qp = hns_roce_v2_query_qp,
6680 	.req_notify_cq = hns_roce_v2_req_notify_cq,
6681 };
6682 
6683 static const struct ib_device_ops hns_roce_v2_dev_srq_ops = {
6684 	.modify_srq = hns_roce_v2_modify_srq,
6685 	.post_srq_recv = hns_roce_v2_post_srq_recv,
6686 	.query_srq = hns_roce_v2_query_srq,
6687 };
6688 
6689 static const struct hns_roce_hw hns_roce_hw_v2 = {
6690 	.cmq_init = hns_roce_v2_cmq_init,
6691 	.cmq_exit = hns_roce_v2_cmq_exit,
6692 	.hw_profile = hns_roce_v2_profile,
6693 	.hw_init = hns_roce_v2_init,
6694 	.hw_exit = hns_roce_v2_exit,
6695 	.post_mbox = v2_post_mbox,
6696 	.poll_mbox_done = v2_poll_mbox_done,
6697 	.chk_mbox_avail = v2_chk_mbox_is_avail,
6698 	.set_gid = hns_roce_v2_set_gid,
6699 	.set_mac = hns_roce_v2_set_mac,
6700 	.write_mtpt = hns_roce_v2_write_mtpt,
6701 	.rereg_write_mtpt = hns_roce_v2_rereg_write_mtpt,
6702 	.frmr_write_mtpt = hns_roce_v2_frmr_write_mtpt,
6703 	.mw_write_mtpt = hns_roce_v2_mw_write_mtpt,
6704 	.write_cqc = hns_roce_v2_write_cqc,
6705 	.set_hem = hns_roce_v2_set_hem,
6706 	.clear_hem = hns_roce_v2_clear_hem,
6707 	.modify_qp = hns_roce_v2_modify_qp,
6708 	.dereg_mr = hns_roce_v2_dereg_mr,
6709 	.qp_flow_control_init = hns_roce_v2_qp_flow_control_init,
6710 	.init_eq = hns_roce_v2_init_eq_table,
6711 	.cleanup_eq = hns_roce_v2_cleanup_eq_table,
6712 	.write_srqc = hns_roce_v2_write_srqc,
6713 	.query_cqc = hns_roce_v2_query_cqc,
6714 	.query_qpc = hns_roce_v2_query_qpc,
6715 	.query_mpt = hns_roce_v2_query_mpt,
6716 	.query_srqc = hns_roce_v2_query_srqc,
6717 	.query_hw_counter = hns_roce_hw_v2_query_counter,
6718 	.hns_roce_dev_ops = &hns_roce_v2_dev_ops,
6719 	.hns_roce_dev_srq_ops = &hns_roce_v2_dev_srq_ops,
6720 };
6721 
6722 static const struct pci_device_id hns_roce_hw_v2_pci_tbl[] = {
6723 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA), 0},
6724 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC), 0},
6725 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA), 0},
6726 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC), 0},
6727 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC), 0},
6728 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_200G_RDMA), 0},
6729 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_RDMA_DCB_PFC_VF),
6730 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
6731 	/* required last entry */
6732 	{0, }
6733 };
6734 
6735 MODULE_DEVICE_TABLE(pci, hns_roce_hw_v2_pci_tbl);
6736 
6737 static void hns_roce_hw_v2_get_cfg(struct hns_roce_dev *hr_dev,
6738 				  struct hnae3_handle *handle)
6739 {
6740 	struct hns_roce_v2_priv *priv = hr_dev->priv;
6741 	const struct pci_device_id *id;
6742 	int i;
6743 
6744 	hr_dev->pci_dev = handle->pdev;
6745 	id = pci_match_id(hns_roce_hw_v2_pci_tbl, hr_dev->pci_dev);
6746 	hr_dev->is_vf = id->driver_data;
6747 	hr_dev->dev = &handle->pdev->dev;
6748 	hr_dev->hw = &hns_roce_hw_v2;
6749 	hr_dev->sdb_offset = ROCEE_DB_SQ_L_0_REG;
6750 	hr_dev->odb_offset = hr_dev->sdb_offset;
6751 
6752 	/* Get info from NIC driver. */
6753 	hr_dev->reg_base = handle->rinfo.roce_io_base;
6754 	hr_dev->mem_base = handle->rinfo.roce_mem_base;
6755 	hr_dev->caps.num_ports = 1;
6756 	hr_dev->iboe.netdevs[0] = handle->rinfo.netdev;
6757 	hr_dev->iboe.phy_port[0] = 0;
6758 
6759 	addrconf_addr_eui48((u8 *)&hr_dev->ib_dev.node_guid,
6760 			    hr_dev->iboe.netdevs[0]->dev_addr);
6761 
6762 	for (i = 0; i < handle->rinfo.num_vectors; i++)
6763 		hr_dev->irq[i] = pci_irq_vector(handle->pdev,
6764 						i + handle->rinfo.base_vector);
6765 
6766 	/* cmd issue mode: 0 is poll, 1 is event */
6767 	hr_dev->cmd_mod = 1;
6768 	hr_dev->loop_idc = 0;
6769 
6770 	hr_dev->reset_cnt = handle->ae_algo->ops->ae_dev_reset_cnt(handle);
6771 	priv->handle = handle;
6772 }
6773 
6774 static int __hns_roce_hw_v2_init_instance(struct hnae3_handle *handle)
6775 {
6776 	struct hns_roce_dev *hr_dev;
6777 	int ret;
6778 
6779 	hr_dev = ib_alloc_device(hns_roce_dev, ib_dev);
6780 	if (!hr_dev)
6781 		return -ENOMEM;
6782 
6783 	hr_dev->priv = kzalloc(sizeof(struct hns_roce_v2_priv), GFP_KERNEL);
6784 	if (!hr_dev->priv) {
6785 		ret = -ENOMEM;
6786 		goto error_failed_kzalloc;
6787 	}
6788 
6789 	hns_roce_hw_v2_get_cfg(hr_dev, handle);
6790 
6791 	ret = hns_roce_init(hr_dev);
6792 	if (ret) {
6793 		dev_err(hr_dev->dev, "RoCE Engine init failed!\n");
6794 		goto error_failed_roce_init;
6795 	}
6796 
6797 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
6798 		ret = free_mr_init(hr_dev);
6799 		if (ret) {
6800 			dev_err(hr_dev->dev, "failed to init free mr!\n");
6801 			goto error_failed_free_mr_init;
6802 		}
6803 	}
6804 
6805 	handle->priv = hr_dev;
6806 
6807 	return 0;
6808 
6809 error_failed_free_mr_init:
6810 	hns_roce_exit(hr_dev);
6811 
6812 error_failed_roce_init:
6813 	kfree(hr_dev->priv);
6814 
6815 error_failed_kzalloc:
6816 	ib_dealloc_device(&hr_dev->ib_dev);
6817 
6818 	return ret;
6819 }
6820 
6821 static void __hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle,
6822 					   bool reset)
6823 {
6824 	struct hns_roce_dev *hr_dev = handle->priv;
6825 
6826 	if (!hr_dev)
6827 		return;
6828 
6829 	handle->priv = NULL;
6830 
6831 	hr_dev->state = HNS_ROCE_DEVICE_STATE_UNINIT;
6832 	hns_roce_handle_device_err(hr_dev);
6833 
6834 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08)
6835 		free_mr_exit(hr_dev);
6836 
6837 	hns_roce_exit(hr_dev);
6838 	kfree(hr_dev->priv);
6839 	ib_dealloc_device(&hr_dev->ib_dev);
6840 }
6841 
6842 static int hns_roce_hw_v2_init_instance(struct hnae3_handle *handle)
6843 {
6844 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
6845 	const struct pci_device_id *id;
6846 	struct device *dev = &handle->pdev->dev;
6847 	int ret;
6848 
6849 	handle->rinfo.instance_state = HNS_ROCE_STATE_INIT;
6850 
6851 	if (ops->ae_dev_resetting(handle) || ops->get_hw_reset_stat(handle)) {
6852 		handle->rinfo.instance_state = HNS_ROCE_STATE_NON_INIT;
6853 		goto reset_chk_err;
6854 	}
6855 
6856 	id = pci_match_id(hns_roce_hw_v2_pci_tbl, handle->pdev);
6857 	if (!id)
6858 		return 0;
6859 
6860 	if (id->driver_data && handle->pdev->revision == PCI_REVISION_ID_HIP08)
6861 		return 0;
6862 
6863 	ret = __hns_roce_hw_v2_init_instance(handle);
6864 	if (ret) {
6865 		handle->rinfo.instance_state = HNS_ROCE_STATE_NON_INIT;
6866 		dev_err(dev, "RoCE instance init failed! ret = %d\n", ret);
6867 		if (ops->ae_dev_resetting(handle) ||
6868 		    ops->get_hw_reset_stat(handle))
6869 			goto reset_chk_err;
6870 		else
6871 			return ret;
6872 	}
6873 
6874 	handle->rinfo.instance_state = HNS_ROCE_STATE_INITED;
6875 
6876 	return 0;
6877 
6878 reset_chk_err:
6879 	dev_err(dev, "Device is busy in resetting state.\n"
6880 		     "please retry later.\n");
6881 
6882 	return -EBUSY;
6883 }
6884 
6885 static void hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle,
6886 					   bool reset)
6887 {
6888 	if (handle->rinfo.instance_state != HNS_ROCE_STATE_INITED)
6889 		return;
6890 
6891 	handle->rinfo.instance_state = HNS_ROCE_STATE_UNINIT;
6892 
6893 	__hns_roce_hw_v2_uninit_instance(handle, reset);
6894 
6895 	handle->rinfo.instance_state = HNS_ROCE_STATE_NON_INIT;
6896 }
6897 static int hns_roce_hw_v2_reset_notify_down(struct hnae3_handle *handle)
6898 {
6899 	struct hns_roce_dev *hr_dev;
6900 
6901 	if (handle->rinfo.instance_state != HNS_ROCE_STATE_INITED) {
6902 		set_bit(HNS_ROCE_RST_DIRECT_RETURN, &handle->rinfo.state);
6903 		return 0;
6904 	}
6905 
6906 	handle->rinfo.reset_state = HNS_ROCE_STATE_RST_DOWN;
6907 	clear_bit(HNS_ROCE_RST_DIRECT_RETURN, &handle->rinfo.state);
6908 
6909 	hr_dev = handle->priv;
6910 	if (!hr_dev)
6911 		return 0;
6912 
6913 	hr_dev->active = false;
6914 	hr_dev->dis_db = true;
6915 	hr_dev->state = HNS_ROCE_DEVICE_STATE_RST_DOWN;
6916 
6917 	return 0;
6918 }
6919 
6920 static int hns_roce_hw_v2_reset_notify_init(struct hnae3_handle *handle)
6921 {
6922 	struct device *dev = &handle->pdev->dev;
6923 	int ret;
6924 
6925 	if (test_and_clear_bit(HNS_ROCE_RST_DIRECT_RETURN,
6926 			       &handle->rinfo.state)) {
6927 		handle->rinfo.reset_state = HNS_ROCE_STATE_RST_INITED;
6928 		return 0;
6929 	}
6930 
6931 	handle->rinfo.reset_state = HNS_ROCE_STATE_RST_INIT;
6932 
6933 	dev_info(&handle->pdev->dev, "In reset process RoCE client reinit.\n");
6934 	ret = __hns_roce_hw_v2_init_instance(handle);
6935 	if (ret) {
6936 		/* when reset notify type is HNAE3_INIT_CLIENT In reset notify
6937 		 * callback function, RoCE Engine reinitialize. If RoCE reinit
6938 		 * failed, we should inform NIC driver.
6939 		 */
6940 		handle->priv = NULL;
6941 		dev_err(dev, "In reset process RoCE reinit failed %d.\n", ret);
6942 	} else {
6943 		handle->rinfo.reset_state = HNS_ROCE_STATE_RST_INITED;
6944 		dev_info(dev, "reset done, RoCE client reinit finished.\n");
6945 	}
6946 
6947 	return ret;
6948 }
6949 
6950 static int hns_roce_hw_v2_reset_notify_uninit(struct hnae3_handle *handle)
6951 {
6952 	if (test_bit(HNS_ROCE_RST_DIRECT_RETURN, &handle->rinfo.state))
6953 		return 0;
6954 
6955 	handle->rinfo.reset_state = HNS_ROCE_STATE_RST_UNINIT;
6956 	dev_info(&handle->pdev->dev, "In reset process RoCE client uninit.\n");
6957 	msleep(HNS_ROCE_V2_HW_RST_UNINT_DELAY);
6958 	__hns_roce_hw_v2_uninit_instance(handle, false);
6959 
6960 	return 0;
6961 }
6962 
6963 static int hns_roce_hw_v2_reset_notify(struct hnae3_handle *handle,
6964 				       enum hnae3_reset_notify_type type)
6965 {
6966 	int ret = 0;
6967 
6968 	switch (type) {
6969 	case HNAE3_DOWN_CLIENT:
6970 		ret = hns_roce_hw_v2_reset_notify_down(handle);
6971 		break;
6972 	case HNAE3_INIT_CLIENT:
6973 		ret = hns_roce_hw_v2_reset_notify_init(handle);
6974 		break;
6975 	case HNAE3_UNINIT_CLIENT:
6976 		ret = hns_roce_hw_v2_reset_notify_uninit(handle);
6977 		break;
6978 	default:
6979 		break;
6980 	}
6981 
6982 	return ret;
6983 }
6984 
6985 static const struct hnae3_client_ops hns_roce_hw_v2_ops = {
6986 	.init_instance = hns_roce_hw_v2_init_instance,
6987 	.uninit_instance = hns_roce_hw_v2_uninit_instance,
6988 	.reset_notify = hns_roce_hw_v2_reset_notify,
6989 };
6990 
6991 static struct hnae3_client hns_roce_hw_v2_client = {
6992 	.name = "hns_roce_hw_v2",
6993 	.type = HNAE3_CLIENT_ROCE,
6994 	.ops = &hns_roce_hw_v2_ops,
6995 };
6996 
6997 static int __init hns_roce_hw_v2_init(void)
6998 {
6999 	hns_roce_init_debugfs();
7000 	return hnae3_register_client(&hns_roce_hw_v2_client);
7001 }
7002 
7003 static void __exit hns_roce_hw_v2_exit(void)
7004 {
7005 	hnae3_unregister_client(&hns_roce_hw_v2_client);
7006 	hns_roce_cleanup_debugfs();
7007 }
7008 
7009 module_init(hns_roce_hw_v2_init);
7010 module_exit(hns_roce_hw_v2_exit);
7011 
7012 MODULE_LICENSE("Dual BSD/GPL");
7013 MODULE_AUTHOR("Wei Hu <xavier.huwei@huawei.com>");
7014 MODULE_AUTHOR("Lijun Ou <oulijun@huawei.com>");
7015 MODULE_AUTHOR("Shaobo Xu <xushaobo2@huawei.com>");
7016 MODULE_DESCRIPTION("Hisilicon Hip08 Family RoCE Driver");
7017